Skip to content

Listings

All listings: OpenRouter

Every listing for the models on page 6 of 26 of OpenRouter, including each model's other versions and routes. Back to OpenRouter

Sold by OpenRouter

117 listings on this page sold by OpenRouter.

Select a score for details.

About these columns
llminfo IQ
llminfo IQ (capability index). How llminfo IQ is computed
Coding IQ
llminfo Coding IQ (coding capability, its own scale). How Coding IQ is computed
Value
Value for money. 50 means as expected for its results. Above 50, cheaper per task. How Value for money works
Yap
Tokens used per coding task vs. models with the same results. Includes agents that write code. Chattiness on maths doesn't predict it. How the Yap score works
Modelllminfo IQCoding IQInput $/1MOutput $/1MMax Input TokensValueYapEpoch Capabilities Index
GLM 5V Turbo
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
126.8
From 68 benchmarks. Plenty of data behind it.
102 ±22
Coding IQ 101.9, 95% interval 80–124. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$1.20cache read $0.24$4.00203K
52.2
Fair value (52.2). Costs per task about what its benchmark results predict. Little data behind it, so treat it as a rough estimate.
Not measuredNot measured
GLM 5V Turbo
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
fp8
Inference by Z.ai
126.8
From 68 benchmarks. Plenty of data behind it.
102 ±22
Coding IQ 101.9, 95% interval 80–124. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$1.20cache read $0.24$4.00203K
52.2
Fair value (52.2). Costs per task about what its benchmark results predict. Little data behind it, so treat it as a rough estimate.
Not measuredNot measured
MiniMax M3
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
126.8
From 332 benchmarks. Plenty of data behind it.
107 ±9
Coding IQ 107.2, 95% interval 98–117. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.30cache read $0.06$1.20524K
50.9
Fair value (50.9). Costs per task about what its benchmark results predict. Plenty of data behind it.
71 ±8
Chatty (71 ±8). Generates more tokens than models with similar coding results. Measured at default effort. A fair amount of data behind it.
147.8
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
MiniMax M3
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
minimax/minimax-m3:batch · batch
126.8
From 332 benchmarks. Plenty of data behind it.
107 ±9
Coding IQ 107.2, 95% interval 98–117. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.224cache read $0.0448$0.88524K
50.9
Fair value (50.9). Costs per task about what its benchmark results predict. Plenty of data behind it.
71 ±8
Chatty (71 ±8). Generates more tokens than models with similar coding results. Measured at default effort. A fair amount of data behind it.
147.8
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
MiniMax M3
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
fp4
Inference by Coreweave
126.8
From 332 benchmarks. Plenty of data behind it.
107 ±9
Coding IQ 107.2, 95% interval 98–117. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.23cache read $0.05$0.96262K
50.9
Fair value (50.9). Costs per task about what its benchmark results predict. Plenty of data behind it.
71 ±8
Chatty (71 ±8). Generates more tokens than models with similar coding results. Measured at default effort. A fair amount of data behind it.
147.8
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
MiniMax M3
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
minimax/minimax-m3:batch · batch+fp8 · fp8
Inference by DeepInfra
126.8
From 332 benchmarks. Plenty of data behind it.
107 ±9
Coding IQ 107.2, 95% interval 98–117. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.224cache read $0.0448$0.88524K
50.9
Fair value (50.9). Costs per task about what its benchmark results predict. Plenty of data behind it.
71 ±8
Chatty (71 ±8). Generates more tokens than models with similar coding results. Measured at default effort. A fair amount of data behind it.
147.8
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
MiniMax M3
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
fp8
Inference by DeepInfra
126.8
From 332 benchmarks. Plenty of data behind it.
107 ±9
Coding IQ 107.2, 95% interval 98–117. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.28cache read $0.056$1.10524K
50.9
Fair value (50.9). Costs per task about what its benchmark results predict. Plenty of data behind it.
71 ±8
Chatty (71 ±8). Generates more tokens than models with similar coding results. Measured at default effort. A fair amount of data behind it.
147.8
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
MiniMax M3
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
fp8/flex · fp8
Inference by DeepInfra
126.8
From 332 benchmarks. Plenty of data behind it.
107 ±9
Coding IQ 107.2, 95% interval 98–117. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.224cache read $0.0448$0.88524K
50.9
Fair value (50.9). Costs per task about what its benchmark results predict. Plenty of data behind it.
71 ±8
Chatty (71 ±8). Generates more tokens than models with similar coding results. Measured at default effort. A fair amount of data behind it.
147.8
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
MiniMax M3
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
fp8
Inference by Featherless AI
126.8
From 332 benchmarks. Plenty of data behind it.
107 ±9
Coding IQ 107.2, 95% interval 98–117. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.55cache read $0.06$2.20262K
50.9
Fair value (50.9). Costs per task about what its benchmark results predict. Plenty of data behind it.
71 ±8
Chatty (71 ±8). Generates more tokens than models with similar coding results. Measured at default effort. A fair amount of data behind it.
147.8
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
MiniMax M3
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
fast · fp8
Inference by Parasail
126.8
From 332 benchmarks. Plenty of data behind it.
107 ±9
Coding IQ 107.2, 95% interval 98–117. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.45cache read $0.09$1.801M
50.9
Fair value (50.9). Costs per task about what its benchmark results predict. Plenty of data behind it.
71 ±8
Chatty (71 ±8). Generates more tokens than models with similar coding results. Measured at default effort. A fair amount of data behind it.
147.8
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
MiniMax M3
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
fp8
Inference by Parasail
126.8
From 332 benchmarks. Plenty of data behind it.
107 ±9
Coding IQ 107.2, 95% interval 98–117. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.30cache read $0.06$1.201M
50.9
Fair value (50.9). Costs per task about what its benchmark results predict. Plenty of data behind it.
71 ±8
Chatty (71 ±8). Generates more tokens than models with similar coding results. Measured at default effort. A fair amount of data behind it.
147.8
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
MiniMax M3
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
fp8
Inference by Venice
126.8
From 332 benchmarks. Plenty of data behind it.
107 ±9
Coding IQ 107.2, 95% interval 98–117. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.30cache read $0.06$1.20524K
50.9
Fair value (50.9). Costs per task about what its benchmark results predict. Plenty of data behind it.
71 ±8
Chatty (71 ±8). Generates more tokens than models with similar coding results. Measured at default effort. A fair amount of data behind it.
147.8
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Qwen3.5 Plus
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
Newer version
Newer versionQwen3.7 PlusWe compared release dates and version names in our catalogue.
qwen/qwen3.5-plus-20260420 · 20260420
126.8
From 53 benchmarks. Plenty of data behind it.
104 ±7
Coding IQ 103.9, 95% interval 97–111. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.30$1.80cache write $0.3751M
51.4
Fair value (51.4). Costs per task about what its benchmark results predict. Plenty of data behind it.
46 ±9
Typical (46 ±9). Generates about as many tokens as models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
Not measured
Qwen3.5 Plus
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
Newer version
Newer versionQwen3.7 PlusWe compared release dates and version names in our catalogue.
qwen/qwen3.5-plus-20260420 · 20260420
Inference by Alibaba Cloud
126.8
From 53 benchmarks. Plenty of data behind it.
104 ±7
Coding IQ 103.9, 95% interval 97–111. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.30$1.80cache write $0.375984K
51.4
Fair value (51.4). Costs per task about what its benchmark results predict. Plenty of data behind it.
46 ±9
Typical (46 ±9). Generates about as many tokens as models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
Not measured
Qwen3.5 122B-A10B
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
126.7
From 287 benchmarks. Plenty of data behind it.
99 ±7
Coding IQ 99.3, 95% interval 92–107. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.26$2.08262K
63.8
Good value (63.8). Cheaper per task than its benchmark results predict. A fair amount of data behind it.
Not measuredNot measured
Qwen3.5 122B-A10B
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
Inference by Alibaba Cloud
126.7
From 287 benchmarks. Plenty of data behind it.
99 ±7
Coding IQ 99.3, 95% interval 92–107. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.26$2.08258K
63.8
Good value (63.8). Cheaper per task than its benchmark results predict. A fair amount of data behind it.
Not measuredNot measured
Qwen3.5 122B-A10B
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
bf16
Inference by Novita AI
126.7
From 287 benchmarks. Plenty of data behind it.
99 ±7
Coding IQ 99.3, 95% interval 92–107. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.40$3.20262K
63.8
Good value (63.8). Cheaper per task than its benchmark results predict. A fair amount of data behind it.
Not measuredNot measured
Qwen3.5 122B-A10B
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
fp8
Inference by SiliconFlow
126.7
From 287 benchmarks. Plenty of data behind it.
99 ±7
Coding IQ 99.3, 95% interval 92–107. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.26$2.08262K
63.8
Good value (63.8). Cheaper per task than its benchmark results predict. A fair amount of data behind it.
Not measuredNot measured
Particle 1
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
≈ 126
From 3 benchmarks. Little data behind it, so treat it as a rough estimate.
Not enough data yet$0.07cache read $0.03$0.501MNot enough data yetNot measuredNot measured
Particle 1
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
particle-ai/particle-1:free · free
≈ 126
From 3 benchmarks. Little data behind it, so treat it as a rough estimate.
Not enough data yet$0.00$0.001MNot enough data yetNot measuredNot measured
Particle 1
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
particle-ai/particle-1:free · free+us
Inference by Particle AI
≈ 126
From 3 benchmarks. Little data behind it, so treat it as a rough estimate.
Not enough data yet$0.00$0.001MNot enough data yetNot measuredNot measured
Particle 1
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
us
Inference by Particle AI
≈ 126
From 3 benchmarks. Little data behind it, so treat it as a rough estimate.
Not enough data yet$0.07cache read $0.03$0.501MNot enough data yetNot measuredNot measured
Particle 1
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
us/flex
Inference by Particle AI
≈ 126
From 3 benchmarks. Little data behind it, so treat it as a rough estimate.
Not enough data yet$0.035cache read $0.015$0.251MNot enough data yetNot measuredNot measured
Nemotron 3.5 Super VL
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
125.9
From 18 benchmarks. A fair amount of data behind it.
Not enough data yet$0.15cache read $0.02$0.471M
60.2
Good value (60.2). Cheaper per task than its benchmark results predict. A fair amount of data behind it.
54 ±9
Typical (54 ±9). Generates about as many tokens as models with similar coding results. Measured with reasoning off. Little data behind it, so treat it as a rough estimate.
Not measured
Nemotron 3.5 Super VL
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
nvidia/nemotron-3.5-super-vl:free · free
125.9
From 18 benchmarks. A fair amount of data behind it.
Not enough data yet$0.00$0.001M
60.2
Good value (60.2). Cheaper per task than its benchmark results predict. A fair amount of data behind it.
54 ±9
Typical (54 ±9). Generates about as many tokens as models with similar coding results. Measured with reasoning off. Little data behind it, so treat it as a rough estimate.
Not measured
Nemotron 3.5 Super VL
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
fp4
Inference by Baseten
125.9
From 18 benchmarks. A fair amount of data behind it.
Not enough data yet$0.20cache read $0.04$0.501M
60.2
Good value (60.2). Cheaper per task than its benchmark results predict. A fair amount of data behind it.
54 ±9
Typical (54 ±9). Generates about as many tokens as models with similar coding results. Measured with reasoning off. Little data behind it, so treat it as a rough estimate.
Not measured
Nemotron 3.5 Super VL
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
nvfp4
Inference by Coreweave
125.9
From 18 benchmarks. A fair amount of data behind it.
Not enough data yet$0.15cache read $0.02$0.471M
60.2
Good value (60.2). Cheaper per task than its benchmark results predict. A fair amount of data behind it.
54 ±9
Typical (54 ±9). Generates about as many tokens as models with similar coding results. Measured with reasoning off. Little data behind it, so treat it as a rough estimate.
Not measured
Nemotron 3.5 Super VL
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
bf16
Inference by Crusoe
125.9
From 18 benchmarks. A fair amount of data behind it.
Not enough data yet$0.10cache read $0.08$0.401M
60.2
Good value (60.2). Cheaper per task than its benchmark results predict. A fair amount of data behind it.
54 ±9
Typical (54 ±9). Generates about as many tokens as models with similar coding results. Measured with reasoning off. Little data behind it, so treat it as a rough estimate.
Not measured
Nemotron 3.5 Super VL
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
nvfp4
Inference by Darkbloom
125.9
From 18 benchmarks. A fair amount of data behind it.
Not enough data yet$0.06cache read $0.01$0.35262K
60.2
Good value (60.2). Cheaper per task than its benchmark results predict. A fair amount of data behind it.
54 ±9
Typical (54 ±9). Generates about as many tokens as models with similar coding results. Measured with reasoning off. Little data behind it, so treat it as a rough estimate.
Not measured
Nemotron 3.5 Super VL
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
fp4
Inference by DeepInfra
125.9
From 18 benchmarks. A fair amount of data behind it.
Not enough data yet$0.15cache read $0.02$0.50262K
60.2
Good value (60.2). Cheaper per task than its benchmark results predict. A fair amount of data behind it.
54 ±9
Typical (54 ±9). Generates about as many tokens as models with similar coding results. Measured with reasoning off. Little data behind it, so treat it as a rough estimate.
Not measured
Nemotron 3.5 Super VL
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
nvfp4
Inference by Ionet
125.9
From 18 benchmarks. A fair amount of data behind it.
Not enough data yet$0.09cache read $0.012$0.30262K
60.2
Good value (60.2). Cheaper per task than its benchmark results predict. A fair amount of data behind it.
54 ±9
Typical (54 ±9). Generates about as many tokens as models with similar coding results. Measured with reasoning off. Little data behind it, so treat it as a rough estimate.
Not measured
Nemotron 3.5 Super VL
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
fp4
Inference by Nebius AI Studio
125.9
From 18 benchmarks. A fair amount of data behind it.
Not enough data yet$0.30cache read $0.06$0.901M
60.2
Good value (60.2). Cheaper per task than its benchmark results predict. A fair amount of data behind it.
54 ±9
Typical (54 ±9). Generates about as many tokens as models with similar coding results. Measured with reasoning off. Little data behind it, so treat it as a rough estimate.
Not measured
Nemotron 3.5 Super VL
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
nvidia/nemotron-3.5-super-vl:free · free
Inference by NVIDIA NIM
125.9
From 18 benchmarks. A fair amount of data behind it.
Not enough data yet$0.00$0.00869K
60.2
Good value (60.2). Cheaper per task than its benchmark results predict. A fair amount of data behind it.
54 ±9
Typical (54 ±9). Generates about as many tokens as models with similar coding results. Measured with reasoning off. Little data behind it, so treat it as a rough estimate.
Not measured
Nemotron 3.5 Super VL
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
nvfp4
Inference by PrimaLabs
125.9
From 18 benchmarks. A fair amount of data behind it.
Not enough data yet$0.15cache read $0.02$0.501M
60.2
Good value (60.2). Cheaper per task than its benchmark results predict. A fair amount of data behind it.
54 ±9
Typical (54 ±9). Generates about as many tokens as models with similar coding results. Measured with reasoning off. Little data behind it, so treat it as a rough estimate.
Not measured
Nemotron 3.5 Super VL
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
nvfp4
Inference by Primeintellect
125.9
From 18 benchmarks. A fair amount of data behind it.
Not enough data yet$0.15cache read $0.02$0.451M
60.2
Good value (60.2). Cheaper per task than its benchmark results predict. A fair amount of data behind it.
54 ±9
Typical (54 ±9). Generates about as many tokens as models with similar coding results. Measured with reasoning off. Little data behind it, so treat it as a rough estimate.
Not measured
Nemotron 3.5 Super VL
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
Inference by Together AI
125.9
From 18 benchmarks. A fair amount of data behind it.
Not enough data yet$0.15cache read $0.02$0.50262K
60.2
Good value (60.2). Cheaper per task than its benchmark results predict. A fair amount of data behind it.
54 ±9
Typical (54 ±9). Generates about as many tokens as models with similar coding results. Measured with reasoning off. Little data behind it, so treat it as a rough estimate.
Not measured
Nemotron 3.5 Super VL
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
fp4
Inference by Venice
125.9
From 18 benchmarks. A fair amount of data behind it.
Not enough data yet$0.20cache read $0.04$0.50262K
60.2
Good value (60.2). Cheaper per task than its benchmark results predict. A fair amount of data behind it.
54 ±9
Typical (54 ±9). Generates about as many tokens as models with similar coding results. Measured with reasoning off. Little data behind it, so treat it as a rough estimate.
Not measured
Step 3.5 Flash
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
Newer version
Newer versionStep 3.7 FlashWe compared release dates and version names in our catalogue.
125.8
From 118 benchmarks. Plenty of data behind it.
106 ±10
Coding IQ 106.2, 95% interval 96–116. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.09cache read $0.02$0.30Not published by the provider
72.1
Good value (72.1). Cheaper per task than its benchmark results predict. Plenty of data behind it.
Not measuredNot measured
Ling 3.0 Flash
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
Newer version
Newer versionLing 3.2 FlashWe compared the version names in our catalogue.
125.6
From 52 benchmarks. Plenty of data behind it.
Not enough data yet$0.021cache read $0.0042$0.063262K
64.2
Good value (64.2). Cheaper per task than its benchmark results predict. Little data behind it, so treat it as a rough estimate.
Not measuredNot measured
Ling 3.0 Flash
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
Newer version
Newer versionLing 3.2 FlashWe compared the version names in our catalogue.
fp4
Inference by DeepInfra
125.6
From 52 benchmarks. Plenty of data behind it.
Not enough data yet$0.06cache read $0.012$0.18131K
64.2
Good value (64.2). Cheaper per task than its benchmark results predict. Little data behind it, so treat it as a rough estimate.
Not measuredNot measured
Ling 3.0 Flash
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
Newer version
Newer versionLing 3.2 FlashWe compared the version names in our catalogue.
Inference by Novita AI
125.6
From 52 benchmarks. Plenty of data behind it.
Not enough data yet$0.021cache read $0.0042$0.063262K
64.2
Good value (64.2). Cheaper per task than its benchmark results predict. Little data behind it, so treat it as a rough estimate.
Not measuredNot measured
Qwen3.5 27B
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
Newer version
Newer versionQwen 3.8 27BWe compared release dates and version names in our catalogue.
125.6
From 331 benchmarks. Plenty of data behind it.
101 ±7
Coding IQ 100.7, 95% interval 94–107. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.26$2.60262K
42.4
Poor value (42.4). Costs more per task than its benchmark results predict. Plenty of data behind it.
56 ±9
Chatty (56 ±9). Generates more tokens than models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
Not measured
Qwen3.5 27B
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
Newer version
Newer versionQwen 3.8 27BWe compared release dates and version names in our catalogue.
Inference by Alibaba Cloud
125.6
From 331 benchmarks. Plenty of data behind it.
101 ±7
Coding IQ 100.7, 95% interval 94–107. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.195$1.56258K
42.4
Poor value (42.4). Costs more per task than its benchmark results predict. Plenty of data behind it.
56 ±9
Chatty (56 ±9). Generates more tokens than models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
Not measured
Qwen3.5 27B
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
Newer version
Newer versionQwen 3.8 27BWe compared release dates and version names in our catalogue.
fp8
Inference by Chutes
125.6
From 331 benchmarks. Plenty of data behind it.
101 ±7
Coding IQ 100.7, 95% interval 94–107. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.20cache read $0.02$1.75262K
42.4
Poor value (42.4). Costs more per task than its benchmark results predict. Plenty of data behind it.
56 ±9
Chatty (56 ±9). Generates more tokens than models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
Not measured
Qwen3.5 27B
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
Newer version
Newer versionQwen 3.8 27BWe compared release dates and version names in our catalogue.
fp8
Inference by DeepInfra
125.6
From 331 benchmarks. Plenty of data behind it.
101 ±7
Coding IQ 100.7, 95% interval 94–107. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.26$2.60262K
42.4
Poor value (42.4). Costs more per task than its benchmark results predict. Plenty of data behind it.
56 ±9
Chatty (56 ±9). Generates more tokens than models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
Not measured
Qwen3.5 27B
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
Newer version
Newer versionQwen 3.8 27BWe compared release dates and version names in our catalogue.
bf16
Inference by Novita AI
125.6
From 331 benchmarks. Plenty of data behind it.
101 ±7
Coding IQ 100.7, 95% interval 94–107. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.30$2.40262K
42.4
Poor value (42.4). Costs more per task than its benchmark results predict. Plenty of data behind it.
56 ±9
Chatty (56 ±9). Generates more tokens than models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
Not measured
Qwen3.5 27B
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
Newer version
Newer versionQwen 3.8 27BWe compared release dates and version names in our catalogue.
fp8
Inference by SiliconFlow
125.6
From 331 benchmarks. Plenty of data behind it.
101 ±7
Coding IQ 100.7, 95% interval 94–107. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.25$2.00262K
42.4
Poor value (42.4). Costs more per task than its benchmark results predict. Plenty of data behind it.
56 ±9
Chatty (56 ±9). Generates more tokens than models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
Not measured
Solar Mini 4
🇰🇷
Maker headquartered in KR, according to the company's own website. The citation is on the model page.
125.4
From 12 benchmarks. A fair amount of data behind it.
Not enough data yet$0.05cache read $0.005$0.20524KNot enough data yetNot measuredNot measured
Solar Mini 4
🇰🇷
Maker headquartered in KR, according to the company's own website. The citation is on the model page.
Inference by Upstage
125.4
From 12 benchmarks. A fair amount of data behind it.
Not enough data yet$0.05cache read $0.005$0.20524KNot enough data yetNot measuredNot measured
Solar Mini 4
🇰🇷
Maker headquartered in KR, according to the company's own website. The citation is on the model page.
zdr
Inference by Upstage
125.4
From 12 benchmarks. A fair amount of data behind it.
Not enough data yet$0.05cache read $0.005$0.20524KNot enough data yetNot measuredNot measured
Gemini 3.1 Flash Lite
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Newer version
Newer versionGemini 3.5 Flash LiteWe compared release dates and version names in our catalogue.
125.3
From 275 benchmarks. Plenty of data behind it.
97 ±10
Coding IQ 97.2, 95% interval 88–107. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.25cache read $0.025$1.501M
55.5
Good value (55.5). Cheaper per task than its benchmark results predict. Plenty of data behind it.
48 ±9
Typical (48 ±9). Generates about as many tokens as models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
145
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Gemini 3.1 Flash Lite
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Newer version
Newer versionGemini 3.5 Flash LiteWe compared release dates and version names in our catalogue.
google/gemini-3.1-flash-lite:batch · batch
125.3
From 275 benchmarks. Plenty of data behind it.
97 ±10
Coding IQ 97.2, 95% interval 88–107. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.125cache read $0.0125$0.751M
55.5
Good value (55.5). Cheaper per task than its benchmark results predict. Plenty of data behind it.
48 ±9
Typical (48 ±9). Generates about as many tokens as models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
145
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Gemini 3.1 Flash Lite
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Newer version
Newer versionGemini 3.5 Flash LiteWe compared release dates and version names in our catalogue.
Inference by Google
125.3
From 275 benchmarks. Plenty of data behind it.
97 ±10
Coding IQ 97.2, 95% interval 88–107. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.25cache read $0.025$1.501M
55.5
Good value (55.5). Cheaper per task than its benchmark results predict. Plenty of data behind it.
48 ±9
Typical (48 ±9). Generates about as many tokens as models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
145
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Gemini 3.1 Flash Lite
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Newer version
Newer versionGemini 3.5 Flash LiteWe compared release dates and version names in our catalogue.
google/gemini-3.1-flash-lite:batch · batch+global
Inference by Google
125.3
From 275 benchmarks. Plenty of data behind it.
97 ±10
Coding IQ 97.2, 95% interval 88–107. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.125cache read $0.0125$0.751M
55.5
Good value (55.5). Cheaper per task than its benchmark results predict. Plenty of data behind it.
48 ±9
Typical (48 ±9). Generates about as many tokens as models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
145
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Gemini 3.1 Flash Lite
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Newer version
Newer versionGemini 3.5 Flash LiteWe compared release dates and version names in our catalogue.
flex
Inference by Google
125.3
From 275 benchmarks. Plenty of data behind it.
97 ±10
Coding IQ 97.2, 95% interval 88–107. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.125cache read $0.0125$0.751M
55.5
Good value (55.5). Cheaper per task than its benchmark results predict. Plenty of data behind it.
48 ±9
Typical (48 ±9). Generates about as many tokens as models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
145
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Gemini 3.1 Flash Lite
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Newer version
Newer versionGemini 3.5 Flash LiteWe compared release dates and version names in our catalogue.
global
Inference by Google
125.3
From 275 benchmarks. Plenty of data behind it.
97 ±10
Coding IQ 97.2, 95% interval 88–107. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.25cache read $0.025$1.501M
55.5
Good value (55.5). Cheaper per task than its benchmark results predict. Plenty of data behind it.
48 ±9
Typical (48 ±9). Generates about as many tokens as models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
145
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Gemini 3.1 Flash Lite
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Newer version
Newer versionGemini 3.5 Flash LiteWe compared release dates and version names in our catalogue.
global/priority
Inference by Google
125.3
From 275 benchmarks. Plenty of data behind it.
97 ±10
Coding IQ 97.2, 95% interval 88–107. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.45cache read $0.045$2.701M
55.5
Good value (55.5). Cheaper per task than its benchmark results predict. Plenty of data behind it.
48 ±9
Typical (48 ±9). Generates about as many tokens as models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
145
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Gemini 3.1 Flash Lite
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Newer version
Newer versionGemini 3.5 Flash LiteWe compared release dates and version names in our catalogue.
priority
Inference by Google
125.3
From 275 benchmarks. Plenty of data behind it.
97 ±10
Coding IQ 97.2, 95% interval 88–107. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.45cache read $0.045$2.701M
55.5
Good value (55.5). Cheaper per task than its benchmark results predict. Plenty of data behind it.
48 ±9
Typical (48 ±9). Generates about as many tokens as models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
145
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Ring 2.6 1T
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
125.3
From 14 benchmarks. A fair amount of data behind it.
102 ±7
Coding IQ 102.2, 95% interval 95–109. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.075cache read $0.015$0.625Not published by the provider
51.2
Fair value (51.2). Costs per task about what its benchmark results predict. Little data behind it, so treat it as a rough estimate.
40 ±9
Terse (40 ±9). Generates fewer tokens than models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
Not measured
Gemini Omni 1.1 Flash
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Newer version
Newer versionGemini Omni 1.2 FlashWe compared the version names in our catalogue.
≈ 125
From 8 benchmarks. Little data behind it, so treat it as a rough estimate.
Not enough data yet$1.50$9.00131KNot enough data yetNot measuredNot measured
Gemini Omni 1.1 Flash
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Newer version
Newer versionGemini Omni 1.2 FlashWe compared the version names in our catalogue.
Inference by Google
≈ 125
From 8 benchmarks. Little data behind it, so treat it as a rough estimate.
Not enough data yet$1.50$9.00131KNot enough data yetNot measuredNot measured
Grok 4
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
125.2
From 296 benchmarks. Plenty of data behind it.
98 ±13
Coding IQ 98.0, 95% interval 85–111. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$3.00$15.00Not published by the provider
32.7
Poor value (32.7). Costs more per task than its benchmark results predict. Plenty of data behind it.
Not measured
146.9
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Qwen3 Max
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
Newer version
Newer versionQwen4 MaxWe compared release dates and version names in our catalogue.
125.2
From 157 benchmarks. Plenty of data behind it.
96 ±10
Coding IQ 96.3, 95% interval 86–107. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$1.20$6.00Not published by the provider
43.3
Poor value (43.3). Costs more per task than its benchmark results predict. Plenty of data behind it.
41 ±9
Terse (41 ±9). Generates fewer tokens than models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
142.4
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Qwen3 Max
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
Newer version
Newer versionQwen4 MaxWe compared release dates and version names in our catalogue.
qwen/qwen3-max-thinking · thinking
125.2
From 157 benchmarks. Plenty of data behind it.
96 ±10
Coding IQ 96.3, 95% interval 86–107. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.78$3.90Not published by the provider
43.3
Poor value (43.3). Costs more per task than its benchmark results predict. Plenty of data behind it.
41 ±9
Terse (41 ±9). Generates fewer tokens than models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
142.4
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
DeepSeek V3.2
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
Newer version
Newer versionDeepSeek V4We compared the version names in our catalogue.
125.0
From 247 benchmarks. Plenty of data behind it.
101 ±7
Coding IQ 101.2, 95% interval 94–108. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.26cache read $0.13$0.38164K
63.8
Good value (63.8). Cheaper per task than its benchmark results predict. Plenty of data behind it.
Not measured
146.3
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
DeepSeek V3.2
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
Newer version
Newer versionDeepSeek V4We compared the version names in our catalogue.
fp8
Inference by Baidu
125.0
From 247 benchmarks. Plenty of data behind it.
101 ±7
Coding IQ 101.2, 95% interval 94–108. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.28cache read $0.028$0.42131K
63.8
Good value (63.8). Cheaper per task than its benchmark results predict. Plenty of data behind it.
Not measured
146.3
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
DeepSeek V3.2
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
Newer version
Newer versionDeepSeek V4We compared the version names in our catalogue.
fp4
Inference by DeepInfra
125.0
From 247 benchmarks. Plenty of data behind it.
101 ±7
Coding IQ 101.2, 95% interval 94–108. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.26cache read $0.13$0.38164K
63.8
Good value (63.8). Cheaper per task than its benchmark results predict. Plenty of data behind it.
Not measured
146.3
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
DeepSeek V3.2
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
Newer version
Newer versionDeepSeek V4We compared the version names in our catalogue.
Inference by Digitalocean
125.0
From 247 benchmarks. Plenty of data behind it.
101 ±7
Coding IQ 101.2, 95% interval 94–108. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.30cache read $0.09$0.96164K
63.8
Good value (63.8). Cheaper per task than its benchmark results predict. Plenty of data behind it.
Not measured
146.3
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
DeepSeek V3.2
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
Newer version
Newer versionDeepSeek V4We compared the version names in our catalogue.
fp8
Inference by SiliconFlow
125.0
From 247 benchmarks. Plenty of data behind it.
101 ±7
Coding IQ 101.2, 95% interval 94–108. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.259cache read $0.14$0.80164K
63.8
Good value (63.8). Cheaper per task than its benchmark results predict. Plenty of data behind it.
Not measured
146.3
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Gemma 4 26B A4B
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
google/gemma-4-26b-a4b-it
124.9
From 216 benchmarks. Plenty of data behind it.
102 ±8
Coding IQ 101.7, 95% interval 94–110. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.0675cache read $0.0375$0.225262K
56.4
Good value (56.4). Cheaper per task than its benchmark results predict. A fair amount of data behind it.
50 ±9
Typical (50 ±9). Generates about as many tokens as models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
141.6
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Gemma 4 26B A4B
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
google/gemma-4-26b-a4b-it:free · free
124.9
From 216 benchmarks. Plenty of data behind it.
102 ±8
Coding IQ 101.7, 95% interval 94–110. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.00$0.00262K
56.4
Good value (56.4). Cheaper per task than its benchmark results predict. A fair amount of data behind it.
50 ±9
Typical (50 ±9). Generates about as many tokens as models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
141.6
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Gemma 4 26B A4B
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
google/gemma-4-26b-a4b-it · fp4
Inference by Akash
124.9
From 216 benchmarks. Plenty of data behind it.
102 ±8
Coding IQ 101.7, 95% interval 94–110. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.07cache read $0.046$0.24131K
56.4
Good value (56.4). Cheaper per task than its benchmark results predict. A fair amount of data behind it.
50 ±9
Typical (50 ±9). Generates about as many tokens as models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
141.6
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Gemma 4 26B A4B
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
google/gemma-4-26b-a4b-it
Inference by Cloudflare Workers AI
124.9
From 216 benchmarks. Plenty of data behind it.
102 ±8
Coding IQ 101.7, 95% interval 94–110. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.10cache read $0.05$0.30256K
56.4
Good value (56.4). Cheaper per task than its benchmark results predict. A fair amount of data behind it.
50 ±9
Typical (50 ±9). Generates about as many tokens as models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
141.6
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Gemma 4 26B A4B
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
google/gemma-4-26b-a4b-it · bf16
Inference by Coreweave
124.9
From 216 benchmarks. Plenty of data behind it.
102 ±8
Coding IQ 101.7, 95% interval 94–110. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.10cache read $0.05$0.30262K
56.4
Good value (56.4). Cheaper per task than its benchmark results predict. A fair amount of data behind it.
50 ±9
Typical (50 ±9). Generates about as many tokens as models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
141.6
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Gemma 4 26B A4B
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
google/gemma-4-26b-a4b-it · int4
Inference by Darkbloom
124.9
From 216 benchmarks. Plenty of data behind it.
102 ±8
Coding IQ 101.7, 95% interval 94–110. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.05cache read $0.0125$0.19262K
56.4
Good value (56.4). Cheaper per task than its benchmark results predict. A fair amount of data behind it.
50 ±9
Typical (50 ±9). Generates about as many tokens as models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
141.6
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Gemma 4 26B A4B
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
google/gemma-4-26b-a4b-it · fp8
Inference by DeepInfra
124.9
From 216 benchmarks. Plenty of data behind it.
102 ±8
Coding IQ 101.7, 95% interval 94–110. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.07$0.34262K
56.4
Good value (56.4). Cheaper per task than its benchmark results predict. A fair amount of data behind it.
50 ±9
Typical (50 ±9). Generates about as many tokens as models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
141.6
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Gemma 4 26B A4B
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
google/gemma-4-26b-a4b-it · bf16
Inference by Dekallm
124.9
From 216 benchmarks. Plenty of data behind it.
102 ±8
Coding IQ 101.7, 95% interval 94–110. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.06cache read $0.04$0.33262K
56.4
Good value (56.4). Cheaper per task than its benchmark results predict. A fair amount of data behind it.
50 ±9
Typical (50 ±9). Generates about as many tokens as models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
141.6
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Gemma 4 26B A4B
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
google/gemma-4-26b-a4b-it:free · free
Inference by Google
124.9
From 216 benchmarks. Plenty of data behind it.
102 ±8
Coding IQ 101.7, 95% interval 94–110. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.00$0.00262K
56.4
Good value (56.4). Cheaper per task than its benchmark results predict. A fair amount of data behind it.
50 ±9
Typical (50 ±9). Generates about as many tokens as models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
141.6
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Gemma 4 26B A4B
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
google/gemma-4-26b-a4b-it · global
Inference by Google
124.9
From 216 benchmarks. Plenty of data behind it.
102 ±8
Coding IQ 101.7, 95% interval 94–110. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.15$0.60262K
56.4
Good value (56.4). Cheaper per task than its benchmark results predict. A fair amount of data behind it.
50 ±9
Typical (50 ±9). Generates about as many tokens as models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
141.6
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Gemma 4 26B A4B
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
google/gemma-4-26b-a4b-it · bf16
Inference by Ionet
124.9
From 216 benchmarks. Plenty of data behind it.
102 ±8
Coding IQ 101.7, 95% interval 94–110. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.067cache read $0.039$0.29262K
56.4
Good value (56.4). Cheaper per task than its benchmark results predict. A fair amount of data behind it.
50 ±9
Typical (50 ±9). Generates about as many tokens as models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
141.6
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Gemma 4 26B A4B
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
google/gemma-4-26b-a4b-it
Inference by Makora
124.9
From 216 benchmarks. Plenty of data behind it.
102 ±8
Coding IQ 101.7, 95% interval 94–110. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.09cache read $0.049$0.28256K
56.4
Good value (56.4). Cheaper per task than its benchmark results predict. A fair amount of data behind it.
50 ±9
Typical (50 ±9). Generates about as many tokens as models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
141.6
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Gemma 4 26B A4B
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
google/gemma-4-26b-a4b-it · bf16
Inference by Nextbit
124.9
From 216 benchmarks. Plenty of data behind it.
102 ±8
Coding IQ 101.7, 95% interval 94–110. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.0675cache read $0.0375$0.225262K
56.4
Good value (56.4). Cheaper per task than its benchmark results predict. A fair amount of data behind it.
50 ±9
Typical (50 ±9). Generates about as many tokens as models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
141.6
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Gemma 4 26B A4B
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
google/gemma-4-26b-a4b-it · bf16
Inference by Parasail
124.9
From 216 benchmarks. Plenty of data behind it.
102 ±8
Coding IQ 101.7, 95% interval 94–110. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.13cache read $0.05$0.40262K
56.4
Good value (56.4). Cheaper per task than its benchmark results predict. A fair amount of data behind it.
50 ±9
Typical (50 ±9). Generates about as many tokens as models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
141.6
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Gemma 4 26B A4B
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
google/gemma-4-26b-a4b-it · bf16
Inference by Primeintellect
124.9
From 216 benchmarks. Plenty of data behind it.
102 ±8
Coding IQ 101.7, 95% interval 94–110. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.082cache read $0.041$0.245262K
56.4
Good value (56.4). Cheaper per task than its benchmark results predict. A fair amount of data behind it.
50 ±9
Typical (50 ±9). Generates about as many tokens as models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
141.6
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Gemma 4 26B A4B
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
google/gemma-4-26b-a4b-it · fp8
Inference by SiliconFlow
124.9
From 216 benchmarks. Plenty of data behind it.
102 ±8
Coding IQ 101.7, 95% interval 94–110. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.14cache read $0.05$0.40262K
56.4
Good value (56.4). Cheaper per task than its benchmark results predict. A fair amount of data behind it.
50 ±9
Typical (50 ±9). Generates about as many tokens as models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
141.6
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Gemma 4 26B A4B
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
google/gemma-4-26b-a4b-it · bf16
Inference by Venice
124.9
From 216 benchmarks. Plenty of data behind it.
102 ±8
Coding IQ 101.7, 95% interval 94–110. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.13cache read $0.05$0.40256K
56.4
Good value (56.4). Cheaper per task than its benchmark results predict. A fair amount of data behind it.
50 ±9
Typical (50 ±9). Generates about as many tokens as models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
141.6
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
gemini-2.5-flash-preview-09-2025
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
124.7
From 26 benchmarks. A fair amount of data behind it.
Not enough data yet$0.30cache read $0.075$2.50cache write $0.383Not published by the provider
63.6
Good value (63.6). Cheaper per task than its benchmark results predict. Plenty of data behind it.
Not measuredNot measured
GPT-5 mini
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Newer version
Newer versionGPT-5.4 miniWe compared release dates and version names in our catalogue.
124.6
From 388 benchmarks. Plenty of data behind it.
98 ±14
Coding IQ 98.3, 95% interval 85–112. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.25cache read $0.025$2.00400K
55.1
Good value (55.1). Cheaper per task than its benchmark results predict. Plenty of data behind it.
40 ±9
Terse (40 ±9). Generates fewer tokens than models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
146
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
GPT-5 mini
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Newer version
Newer versionGPT-5.4 miniWe compared release dates and version names in our catalogue.
openai/gpt-5-mini:batch · batch
124.6
From 388 benchmarks. Plenty of data behind it.
98 ±14
Coding IQ 98.3, 95% interval 85–112. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.125cache read $0.0125$1.00400K
55.1
Good value (55.1). Cheaper per task than its benchmark results predict. Plenty of data behind it.
40 ±9
Terse (40 ±9). Generates fewer tokens than models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
146
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
GPT-5 mini
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Newer version
Newer versionGPT-5.4 miniWe compared release dates and version names in our catalogue.
Inference by Microsoft Azure
124.6
From 388 benchmarks. Plenty of data behind it.
98 ±14
Coding IQ 98.3, 95% interval 85–112. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.25cache read $0.03$2.00400K
55.1
Good value (55.1). Cheaper per task than its benchmark results predict. Plenty of data behind it.
40 ±9
Terse (40 ±9). Generates fewer tokens than models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
146
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
GPT-5 mini
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Newer version
Newer versionGPT-5.4 miniWe compared release dates and version names in our catalogue.
Inference by OpenAI
124.6
From 388 benchmarks. Plenty of data behind it.
98 ±14
Coding IQ 98.3, 95% interval 85–112. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.25cache read $0.025$2.00272K
55.1
Good value (55.1). Cheaper per task than its benchmark results predict. Plenty of data behind it.
40 ±9
Terse (40 ±9). Generates fewer tokens than models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
146
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
GPT-5 mini
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Newer version
Newer versionGPT-5.4 miniWe compared release dates and version names in our catalogue.
flex
Inference by OpenAI
124.6
From 388 benchmarks. Plenty of data behind it.
98 ±14
Coding IQ 98.3, 95% interval 85–112. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.125cache read $0.0125$1.00272K
55.1
Good value (55.1). Cheaper per task than its benchmark results predict. Plenty of data behind it.
40 ±9
Terse (40 ±9). Generates fewer tokens than models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
146
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Ling 3.0 Flash VL
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
≈ 125
From 3 benchmarks. Little data behind it, so treat it as a rough estimate.
Not enough data yet$0.06cache read $0.012$0.18131K
65.3
Good value (65.3). Cheaper per task than its benchmark results predict. Little data behind it, so treat it as a rough estimate.
Not measuredNot measured
Ling 3.0 Flash VL
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
fp4
Inference by DeepInfra
≈ 125
From 3 benchmarks. Little data behind it, so treat it as a rough estimate.
Not enough data yet$0.06cache read $0.012$0.18131K
65.3
Good value (65.3). Cheaper per task than its benchmark results predict. Little data behind it, so treat it as a rough estimate.
Not measuredNot measured
O1 Pro
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Newer version
Newer versionO3 ProWe compared release dates and version names in our catalogue.
124.4
From 11 benchmarks. Plenty of data behind it.
Not enough data yet$150.00$600.00Not published by the provider
39.3
Poor value (39.3). Costs more per task than its benchmark results predict. Little data behind it, so treat it as a rough estimate.
Not measuredNot measured
Qwen 3 235B A22B Thinking 2507
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
124.3
From 161 benchmarks. Plenty of data behind it.
100 ±8stale
Coding IQ 99.6, 95% interval 91–108. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ. Its newest result is over a year old.
$0.45cache read $0.10$3.50128K
40.6
Poor value (40.6). Costs more per task than its benchmark results predict. A fair amount of data behind it.
Not measuredNot measured
Qwen 3 235B A22B Thinking 2507
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
fp8
Inference by Venice
124.3
From 161 benchmarks. Plenty of data behind it.
100 ±8stale
Coding IQ 99.6, 95% interval 91–108. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ. Its newest result is over a year old.
$0.45$3.50128K
40.6
Poor value (40.6). Costs more per task than its benchmark results predict. A fair amount of data behind it.
Not measuredNot measured
Grok 4.3
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
124.2
From 205 benchmarks. Plenty of data behind it.
101 ±8
Coding IQ 101.2, 95% interval 94–109. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$1.25cache read $0.20$2.501M
39.3
Poor value (39.3). Costs more per task than its benchmark results predict. Plenty of data behind it.
48 ±9
Typical (48 ±9). Generates about as many tokens as models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
Not measured
Grok 4.3
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
x-ai/grok-4.3:batch · batch
124.2
From 205 benchmarks. Plenty of data behind it.
101 ±8
Coding IQ 101.2, 95% interval 94–109. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$1.00cache read $0.16$2.001M
39.3
Poor value (39.3). Costs more per task than its benchmark results predict. Plenty of data behind it.
48 ±9
Typical (48 ±9). Generates about as many tokens as models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
Not measured
Grok 4.3
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Inference by xAI
124.2
From 205 benchmarks. Plenty of data behind it.
101 ±8
Coding IQ 101.2, 95% interval 94–109. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$1.25cache read $0.20$2.501M
39.3
Poor value (39.3). Costs more per task than its benchmark results predict. Plenty of data behind it.
48 ±9
Typical (48 ±9). Generates about as many tokens as models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
Not measured
Grok 4.3
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
x-ai/grok-4.3:batch · batch
Inference by xAI
124.2
From 205 benchmarks. Plenty of data behind it.
101 ±8
Coding IQ 101.2, 95% interval 94–109. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$1.00cache read $0.16$2.001M
39.3
Poor value (39.3). Costs more per task than its benchmark results predict. Plenty of data behind it.
48 ±9
Typical (48 ±9). Generates about as many tokens as models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
Not measured
Grok 4.3
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
priority
Inference by xAI
124.2
From 205 benchmarks. Plenty of data behind it.
101 ±8
Coding IQ 101.2, 95% interval 94–109. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$2.50cache read $0.40$5.001M
39.3
Poor value (39.3). Costs more per task than its benchmark results predict. Plenty of data behind it.
48 ±9
Typical (48 ±9). Generates about as many tokens as models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
Not measured
Grok 4.3
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
zdr
Inference by xAI
124.2
From 205 benchmarks. Plenty of data behind it.
101 ±8
Coding IQ 101.2, 95% interval 94–109. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$1.25cache read $0.20$2.501M
39.3
Poor value (39.3). Costs more per task than its benchmark results predict. Plenty of data behind it.
48 ±9
Typical (48 ±9). Generates about as many tokens as models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
Not measured
Grok 4.3
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
zdr/priority
Inference by xAI
124.2
From 205 benchmarks. Plenty of data behind it.
101 ±8
Coding IQ 101.2, 95% interval 94–109. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$2.50cache read $0.40$5.001M
39.3
Poor value (39.3). Costs more per task than its benchmark results predict. Plenty of data behind it.
48 ±9
Typical (48 ±9). Generates about as many tokens as models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
Not measured
Nemotron 3 Ultra
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
nvidia/nemotron-3-ultra-550b-a55b
124.1
From 172 benchmarks. Plenty of data behind it.
101 ±12
Coding IQ 100.9, 95% interval 89–112. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.50cache read $0.10$2.20262K
33.1
Poor value (33.1). Costs more per task than its benchmark results predict. A fair amount of data behind it.
53 ±9
Typical (53 ±9). Generates about as many tokens as models with similar coding results. Measured with reasoning off. Little data behind it, so treat it as a rough estimate.
147
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Nemotron 3 Ultra
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
nvidia/nemotron-3-ultra-550b-a55b:free · free
124.1
From 172 benchmarks. Plenty of data behind it.
101 ±12
Coding IQ 100.9, 95% interval 89–112. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.00$0.001M
33.1
Poor value (33.1). Costs more per task than its benchmark results predict. A fair amount of data behind it.
53 ±9
Typical (53 ±9). Generates about as many tokens as models with similar coding results. Measured with reasoning off. Little data behind it, so treat it as a rough estimate.
147
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Nemotron 3 Ultra
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
nvidia/nemotron-3-ultra-550b-a55b · fp4/us · fp4
Inference by Baseten
124.1
From 172 benchmarks. Plenty of data behind it.
101 ±12
Coding IQ 100.9, 95% interval 89–112. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.60cache read $0.12$2.40203K
33.1
Poor value (33.1). Costs more per task than its benchmark results predict. A fair amount of data behind it.
53 ±9
Typical (53 ±9). Generates about as many tokens as models with similar coding results. Measured with reasoning off. Little data behind it, so treat it as a rough estimate.
147
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Nemotron 3 Ultra
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
nvidia/nemotron-3-ultra-550b-a55b · bf16
Inference by Coreweave
124.1
From 172 benchmarks. Plenty of data behind it.
101 ±12
Coding IQ 100.9, 95% interval 89–112. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.50cache read $0.10$2.15262K
33.1
Poor value (33.1). Costs more per task than its benchmark results predict. A fair amount of data behind it.
53 ±9
Typical (53 ±9). Generates about as many tokens as models with similar coding results. Measured with reasoning off. Little data behind it, so treat it as a rough estimate.
147
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Nemotron 3 Ultra
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
nvidia/nemotron-3-ultra-550b-a55b · fp4
Inference by DeepInfra
124.1
From 172 benchmarks. Plenty of data behind it.
101 ±12
Coding IQ 100.9, 95% interval 89–112. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.50cache read $0.10$2.20262K
33.1
Poor value (33.1). Costs more per task than its benchmark results predict. A fair amount of data behind it.
53 ±9
Typical (53 ±9). Generates about as many tokens as models with similar coding results. Measured with reasoning off. Little data behind it, so treat it as a rough estimate.
147
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Nemotron 3 Ultra
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
nvidia/nemotron-3-ultra-550b-a55b · fp4/flex · fp4
Inference by DeepInfra
124.1
From 172 benchmarks. Plenty of data behind it.
101 ±12
Coding IQ 100.9, 95% interval 89–112. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.40cache read $0.08$1.76262K
33.1
Poor value (33.1). Costs more per task than its benchmark results predict. A fair amount of data behind it.
53 ±9
Typical (53 ±9). Generates about as many tokens as models with similar coding results. Measured with reasoning off. Little data behind it, so treat it as a rough estimate.
147
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Nemotron 3 Ultra
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
nvidia/nemotron-3-ultra-550b-a55b · fp4/priority · fp4
Inference by DeepInfra
124.1
From 172 benchmarks. Plenty of data behind it.
101 ±12
Coding IQ 100.9, 95% interval 89–112. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.75cache read $0.15$3.30262K
33.1
Poor value (33.1). Costs more per task than its benchmark results predict. A fair amount of data behind it.
53 ±9
Typical (53 ±9). Generates about as many tokens as models with similar coding results. Measured with reasoning off. Little data behind it, so treat it as a rough estimate.
147
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Nemotron 3 Ultra
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
nvidia/nemotron-3-ultra-550b-a55b
Inference by Fireworks AI
124.1
From 172 benchmarks. Plenty of data behind it.
101 ±12
Coding IQ 100.9, 95% interval 89–112. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.60cache read $0.12$2.40262K
33.1
Poor value (33.1). Costs more per task than its benchmark results predict. A fair amount of data behind it.
53 ±9
Typical (53 ±9). Generates about as many tokens as models with similar coding results. Measured with reasoning off. Little data behind it, so treat it as a rough estimate.
147
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Nemotron 3 Ultra
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
nvidia/nemotron-3-ultra-550b-a55b
Inference by Kt
124.1
From 172 benchmarks. Plenty of data behind it.
101 ±12
Coding IQ 100.9, 95% interval 89–112. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.50cache read $0.10$2.20262K
33.1
Poor value (33.1). Costs more per task than its benchmark results predict. A fair amount of data behind it.
53 ±9
Typical (53 ±9). Generates about as many tokens as models with similar coding results. Measured with reasoning off. Little data behind it, so treat it as a rough estimate.
147
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Nemotron 3 Ultra
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
nvidia/nemotron-3-ultra-550b-a55b:free · free
Inference by NVIDIA NIM
124.1
From 172 benchmarks. Plenty of data behind it.
101 ±12
Coding IQ 100.9, 95% interval 89–112. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.00$0.001M
33.1
Poor value (33.1). Costs more per task than its benchmark results predict. A fair amount of data behind it.
53 ±9
Typical (53 ±9). Generates about as many tokens as models with similar coding results. Measured with reasoning off. Little data behind it, so treat it as a rough estimate.
147
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Nemotron 3 Ultra
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
nvidia/nemotron-3-ultra-550b-a55b · fp8
Inference by Venice
124.1
From 172 benchmarks. Plenty of data behind it.
101 ±12
Coding IQ 100.9, 95% interval 89–112. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.625cache read $0.1875$3.125256K
33.1
Poor value (33.1). Costs more per task than its benchmark results predict. A fair amount of data behind it.
53 ±9
Typical (53 ±9). Generates about as many tokens as models with similar coding results. Measured with reasoning off. Little data behind it, so treat it as a rough estimate.
147
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Claude Sonnet 3.7
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Newer version
Newer versionClaude Sonnet 5.5We compared the version names in our catalogue.
claude-3.7-sonnet
124.0
From 423 benchmarks. Plenty of data behind it.
94 ±9
Coding IQ 93.6, 95% interval 85–103. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$3.00cache read $0.30$15.00cache write $3.75Not published by the provider
39.2
Poor value (39.2). Costs more per task than its benchmark results predict. Plenty of data behind it.
49 ±11
Typical (49 ±11). Generates about as many tokens as models with similar coding results. Measured at high effort. Little data behind it, so treat it as a rough estimate.
141
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Claude Sonnet 3.7
🇺🇸
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Newer version
Newer versionClaude Sonnet 5.5We compared the version names in our catalogue.
anthropic/claude-3.7-sonnet:thinking · thinking
124.0
From 423 benchmarks. Plenty of data behind it.
94 ±9
Coding IQ 93.6, 95% interval 85–103. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$3.00$15.00Not published by the provider
39.2
Poor value (39.2). Costs more per task than its benchmark results predict. Plenty of data behind it.
49 ±11
Typical (49 ±11). Generates about as many tokens as models with similar coding results. Measured at high effort. Little data behind it, so treat it as a rough estimate.
141
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API

Data credits: Epoch AI Benchmarking Hub, CC BY 4.0 (licence)

How this data works