Skip to content

Listings

All listings: OpenRouter

Every listing for the models on page 10 of 26 of OpenRouter, including each model's other versions and routes. Back to OpenRouter

Sold by OpenRouter

70 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
Codestral 25.01
πŸ‡«πŸ‡·
Maker headquartered in FR, according to the company's own website. The citation is on the model page.
2501
116.6
From 10 benchmarks. A fair amount of data behind it.
Not enough data yet$0.30$0.90Not published by the provider
37.8
Poor value (37.8). Costs more per task than its benchmark results predict. A fair amount of data behind it.
40 Β±9
Terse (40 Β±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
Codestral 25.01
πŸ‡«πŸ‡·
Maker headquartered in FR, according to the company's own website. The citation is on the model page.
mistralai/codestral-2508 Β· 2508
116.6
From 10 benchmarks. A fair amount of data behind it.
Not enough data yet$0.30cache read $0.03$0.90256K
37.8
Poor value (37.8). Costs more per task than its benchmark results predict. A fair amount of data behind it.
40 Β±9
Terse (40 Β±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
Codestral 25.01
πŸ‡«πŸ‡·
Maker headquartered in FR, according to the company's own website. The citation is on the model page.
mistralai/codestral-2508 Β· 2508
Inference by Mistral AI
116.6
From 10 benchmarks. A fair amount of data behind it.
Not enough data yet$0.30cache read $0.03$0.90256K
37.8
Poor value (37.8). Costs more per task than its benchmark results predict. A fair amount of data behind it.
40 Β±9
Terse (40 Β±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
Mercury 2
Newer version
Newer versionMercury 3We compared release dates and version names in our catalogue.
116.5
From 53 benchmarks. Plenty of data behind it.
93Β Β±13
Coding IQ 93.2, 95% interval 80–106. A fair amount 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$0.75Not published by the provider
56.7
Good value (56.7). Cheaper per task than its benchmark results predict. A fair amount of data behind it.
43 Β±9
Terse (43 Β±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
Mistral Medium 3
πŸ‡«πŸ‡·
Maker headquartered in FR, according to the company's own website. The citation is on the model page.
Newer version
Newer versionMistral Medium 3.5We compared release dates and version names in our catalogue.
116.4
From 50 benchmarks. A fair amount of data behind it.
Not enough data yet$0.44cache read $0.044$2.20131K
51.0
Fair value (51.0). Costs per task about what its benchmark results predict. Little data behind it, so treat it as a rough estimate.
Not measured
132.8
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Mistral Medium 3
πŸ‡«πŸ‡·
Maker headquartered in FR, according to the company's own website. The citation is on the model page.
Newer version
Newer versionMistral Medium 3.5We compared release dates and version names in our catalogue.
eu
Inference by Mistral AI
116.4
From 50 benchmarks. A fair amount of data behind it.
Not enough data yet$0.44cache read $0.044$2.20131K
51.0
Fair value (51.0). Costs per task about what its benchmark results predict. Little data behind it, so treat it as a rough estimate.
Not measured
132.8
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Qwen Plus
πŸ‡¨πŸ‡³
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
qwen/qwen-plus-2025-07-28 Β· 2025-07-28
116.4
From 42 benchmarks. Plenty of data behind it.
Not enough data yet$0.26$0.78Not published by the provider
43.6
Poor value (43.6). Costs more per task than its benchmark results predict. Little data behind it, so treat it as a rough estimate.
Not measuredNot measured
Qwen Plus
πŸ‡¨πŸ‡³
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
qwen/qwen-plus-2025-07-28:thinking Β· thinking+2025-07-28
116.4
From 42 benchmarks. Plenty of data behind it.
Not enough data yet$0.26$0.78cache write $0.325Not published by the provider
43.6
Poor value (43.6). Costs more per task than its benchmark results predict. Little data behind it, so treat it as a rough estimate.
Not measuredNot measured
Qwen Plus
πŸ‡¨πŸ‡³
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
116.4
From 42 benchmarks. Plenty of data behind it.
Not enough data yet$0.26cache read $0.052$0.78cache write $0.3251M
43.6
Poor value (43.6). Costs more per task than its benchmark results predict. Little data behind it, so treat it as a rough estimate.
Not measuredNot measured
Qwen Plus
πŸ‡¨πŸ‡³
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
Inference by Alibaba Cloud
116.4
From 42 benchmarks. Plenty of data behind it.
Not enough data yet$0.26cache read $0.052$0.78cache write $0.325996K
43.6
Poor value (43.6). Costs more per task than its benchmark results predict. Little data behind it, so treat it as a rough estimate.
Not measuredNot measured
DeepSeek V3
πŸ‡¨πŸ‡³
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 V3.1We compared release dates and version names in our catalogue.
deepseek/deepseek-chat-v3-0324:free Β· free+0324
116.3
From 9 benchmarks. A fair amount of data behind it.
Not enough data yetβ€”β€”Not published by the provider
46.7
Poor value (46.7). Costs more per task than its benchmark results predict. A fair amount of data behind it.
Not measuredNot measured
DeepSeek V3
πŸ‡¨πŸ‡³
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 V3.1We compared release dates and version names in our catalogue.
deepseek/deepseek-chat-v3-0324 Β· 0324
116.3
From 9 benchmarks. A fair amount of data behind it.
Not enough data yet$0.25$1.00164K
46.7
Poor value (46.7). Costs more per task than its benchmark results predict. A fair amount of data behind it.
Not measuredNot measured
DeepSeek V3
πŸ‡¨πŸ‡³
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 V3.1We compared release dates and version names in our catalogue.
deepseek/deepseek-chat-v3-0324 Β· fp8+0324 Β· fp8
Inference by SiliconFlow
116.3
From 9 benchmarks. A fair amount of data behind it.
Not enough data yet$0.25$1.00164K
46.7
Poor value (46.7). Costs more per task than its benchmark results predict. A fair amount of data behind it.
Not measuredNot measured
GPT-4 Turbo
πŸ‡ΊπŸ‡Έ
Maker headquartered in US, according to the company's own website. The citation is on the model page.
116.2
From 380 benchmarks. Plenty of data behind it.
83Β Β±10
Coding IQ 83.0, 95% interval 73–93. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$10.00$30.00Not published by the provider
34.4
Poor value (34.4). Costs more per task than its benchmark results predict. Little data behind it, so treat it as a rough estimate.
Not measured
125.8
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
GPT-4 Turbo
πŸ‡ΊπŸ‡Έ
Maker headquartered in US, according to the company's own website. The citation is on the model page.
openai/gpt-4-turbo:batch Β· batch
116.2
From 380 benchmarks. Plenty of data behind it.
83Β Β±10
Coding IQ 83.0, 95% interval 73–93. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$5.00$15.00128K
34.4
Poor value (34.4). Costs more per task than its benchmark results predict. Little data behind it, so treat it as a rough estimate.
Not measured
125.8
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Nemotron 3.5 Lightning
πŸ‡ΊπŸ‡Έ
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.
116.2
From 12 benchmarks. A fair amount of data behind it.
Not enough data yet$0.07cache read $0.04$0.20262K
49.3
Fair value (49.3). Costs per task about what its benchmark results predict. Little data behind it, so treat it as a rough estimate.
Not measuredNot measured
Nemotron 3.5 Lightning
πŸ‡ΊπŸ‡Έ
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-lightning:free Β· free
116.2
From 12 benchmarks. A fair amount of data behind it.
Not enough data yet$0.00$0.001M
49.3
Fair value (49.3). Costs per task about what its benchmark results predict. Little data behind it, so treat it as a rough estimate.
Not measuredNot measured
Nemotron 3.5 Lightning
πŸ‡ΊπŸ‡Έ
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 Coreweave
116.2
From 12 benchmarks. A fair amount of data behind it.
Not enough data yet$0.07cache read $0.04$0.20262K
49.3
Fair value (49.3). Costs per task about what its benchmark results predict. Little data behind it, so treat it as a rough estimate.
Not measuredNot measured
Nemotron 3.5 Lightning
πŸ‡ΊπŸ‡Έ
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.
int4
Inference by Darkbloom
116.2
From 12 benchmarks. A fair amount of data behind it.
Not enough data yet$0.02cache read $0.01$0.096262K
49.3
Fair value (49.3). Costs per task about what its benchmark results predict. Little data behind it, so treat it as a rough estimate.
Not measuredNot measured
Nemotron 3.5 Lightning
πŸ‡ΊπŸ‡Έ
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 DeepInfra
116.2
From 12 benchmarks. A fair amount of data behind it.
Not enough data yet$0.06cache read $0.03$0.16262K
49.3
Fair value (49.3). Costs per task about what its benchmark results predict. Little data behind it, so treat it as a rough estimate.
Not measuredNot measured
Nemotron 3.5 Lightning
πŸ‡ΊπŸ‡Έ
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 Fireworks AI
116.2
From 12 benchmarks. A fair amount of data behind it.
Not enough data yet$0.05cache read $0.01$0.20262K
49.3
Fair value (49.3). Costs per task about what its benchmark results predict. Little data behind it, so treat it as a rough estimate.
Not measuredNot measured
Nemotron 3.5 Lightning
πŸ‡ΊπŸ‡Έ
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 Modelrun
116.2
From 12 benchmarks. A fair amount of data behind it.
Not enough data yet$0.07cache read $0.04$0.201M
49.3
Fair value (49.3). Costs per task about what its benchmark results predict. Little data behind it, so treat it as a rough estimate.
Not measuredNot measured
Nemotron 3.5 Lightning
πŸ‡ΊπŸ‡Έ
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-lightning:free Β· free+nvfp4 Β· nvfp4
Inference by NVIDIA NIM
116.2
From 12 benchmarks. A fair amount of data behind it.
Not enough data yet$0.00$0.001M
49.3
Fair value (49.3). Costs per task about what its benchmark results predict. Little data behind it, so treat it as a rough estimate.
Not measuredNot measured
Qwen3 Coder 30B A3B Instruct
πŸ‡¨πŸ‡³
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.
116.2
From 18 benchmarks. A fair amount of data behind it.
Not enough data yet$0.07$0.28262K
54.3
Good value (54.3). Cheaper per task than its benchmark results predict. Little data behind it, so treat it as a rough estimate.
Not measuredNot measured
Qwen3 Coder 30B A3B Instruct
πŸ‡¨πŸ‡³
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.
us-west-2
Inference by Amazon Web Services
116.2
From 18 benchmarks. A fair amount of data behind it.
Not enough data yet$0.15$0.60262K
54.3
Good value (54.3). Cheaper per task than its benchmark results predict. Little data behind it, so treat it as a rough estimate.
Not measuredNot measured
Qwen3 Coder 30B A3B Instruct
πŸ‡¨πŸ‡³
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
116.2
From 18 benchmarks. A fair amount of data behind it.
Not enough data yet$0.07$0.28262K
54.3
Good value (54.3). Cheaper per task than its benchmark results predict. Little data behind it, so treat it as a rough estimate.
Not measuredNot measured
Gemini 2.0 Flash
πŸ‡ΊπŸ‡Έ
Maker headquartered in US, according to the company's own website. The citation is on the model page.
gemini-2.0-flash-001
116.1
From 211 benchmarks. Plenty of data behind it.
β‰ˆ 85Β Β±14stale
Coding IQ 84.6, 95% interval 71–98. Little data behind it, so treat it as a rough estimate. Uses the model's best published results. It has its own scale, separate from llminfo IQ. Its newest result is over a year old.
$0.10cache read $0.025$0.40cache write $0.1833Not published by the provider
52.1
Fair value (52.1). Costs per task about what its benchmark results predict. Plenty of data behind it.
Not measuredNot measured
GLM 4.7 Flash
πŸ‡¨πŸ‡³
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 versionGLM 5.3 FlashWe compared release dates and version names in our catalogue.
116.1
From 186 benchmarks. Plenty of data behind it.
83Β Β±10
Coding IQ 82.8, 95% interval 73–93. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.0605cache read $0.01$0.40131K
44.1
Poor value (44.1). Costs more per task than its benchmark results predict. A fair amount of data behind it.
Not measuredNot measured
GLM 4.7 Flash
πŸ‡¨πŸ‡³
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 versionGLM 5.3 FlashWe compared release dates and version names in our catalogue.
Inference by Cloudflare Workers AI
116.1
From 186 benchmarks. Plenty of data behind it.
83Β Β±10
Coding IQ 82.8, 95% interval 73–93. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.0605$0.40131K
44.1
Poor value (44.1). Costs more per task than its benchmark results predict. A fair amount of data behind it.
Not measuredNot measured
GLM 4.5V
πŸ‡¨πŸ‡³
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.
115.8
From 62 benchmarks. A fair amount of data behind it.
Not enough data yet$0.60cache read $0.11$1.8066K
36.7
Poor value (36.7). Costs more per task than its benchmark results predict. Plenty of data behind it.
Not measuredNot measured
GLM 4.5V
πŸ‡¨πŸ‡³
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 Novita AI
115.8
From 62 benchmarks. A fair amount of data behind it.
Not enough data yet$0.60cache read $0.11$1.8066K
36.7
Poor value (36.7). Costs more per task than its benchmark results predict. Plenty of data behind it.
Not measuredNot measured
GLM 4.5V
πŸ‡¨πŸ‡³
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 Z.ai
115.8
From 62 benchmarks. A fair amount of data behind it.
Not enough data yet$0.60cache read $0.11$1.8066K
36.7
Poor value (36.7). Costs more per task than its benchmark results predict. Plenty of data behind it.
Not measuredNot measured
QwQ 32B
πŸ‡¨πŸ‡³
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.
115.8
From 144 benchmarks. Plenty of data behind it.
86Β Β±11
Coding IQ 86.2, 95% interval 76–97. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.15$0.20Not published by the provider
44.6
Poor value (44.6). Costs more per task than its benchmark results predict. A fair amount of data behind it.
Not measured
136.4
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Gemini 3.1 Flash Image
πŸ‡ΊπŸ‡Έ
Maker headquartered in US, according to the company's own website. The citation is on the model page.
115.7
From 40 benchmarks. A fair amount of data behind it.
Not enough data yet$0.50$3.00131KNot enough data yetNot measuredNot measured
Gemini 3.1 Flash Image
πŸ‡ΊπŸ‡Έ
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Inference by Google
115.7
From 40 benchmarks. A fair amount of data behind it.
Not enough data yet$0.50$3.0066KNot enough data yetNot measuredNot measured
Gemini 3.1 Flash Image
πŸ‡ΊπŸ‡Έ
Maker headquartered in US, according to the company's own website. The citation is on the model page.
global
Inference by Google
115.7
From 40 benchmarks. A fair amount of data behind it.
Not enough data yet$0.50$3.00131KNot enough data yetNot measuredNot measured
Ministral 14B
πŸ‡«πŸ‡·
Maker headquartered in FR, 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.
mistralai/ministral-14b-2512 Β· 2512
115.6
From 33 benchmarks. A fair amount of data behind it.
Not enough data yet$0.20cache read $0.02$0.20262KNot enough data yetNot measuredNot measured
Ministral 14B
πŸ‡«πŸ‡·
Maker headquartered in FR, 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.
mistralai/ministral-14b-2512 Β· eu+2512
Inference by Mistral AI
115.6
From 33 benchmarks. A fair amount of data behind it.
Not enough data yet$0.22cache read $0.022$0.22262KNot enough data yetNot measuredNot measured
Ministral 14B
πŸ‡«πŸ‡·
Maker headquartered in FR, 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.
mistralai/ministral-14b-2512 Β· zdr+2512
Inference by Mistral AI
115.6
From 33 benchmarks. A fair amount of data behind it.
Not enough data yet$0.20cache read $0.02$0.20262KNot enough data yetNot measuredNot measured
Qwen3 32B
πŸ‡¨πŸ‡³
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.
115.6
From 234 benchmarks. Plenty of data behind it.
85Β Β±9
Coding IQ 84.8, 95% interval 76–93. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.08$0.2841K
35.8
Poor value (35.8). Costs more per task than its benchmark results predict. A fair amount of data behind it.
Not measured
136.9
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Qwen3 32B
πŸ‡¨πŸ‡³
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 DeepInfra
115.6
From 234 benchmarks. Plenty of data behind it.
85Β Β±9
Coding IQ 84.8, 95% interval 76–93. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.08$0.2841K
35.8
Poor value (35.8). Costs more per task than its benchmark results predict. A fair amount of data behind it.
Not measured
136.9
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Qwen3 32B
πŸ‡¨πŸ‡³
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/flex Β· fp8
Inference by DeepInfra
115.6
From 234 benchmarks. Plenty of data behind it.
85Β Β±9
Coding IQ 84.8, 95% interval 76–93. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.064$0.22441K
35.8
Poor value (35.8). Costs more per task than its benchmark results predict. A fair amount of data behind it.
Not measured
136.9
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Qwen3 32B
πŸ‡¨πŸ‡³
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/priority Β· fp8
Inference by DeepInfra
115.6
From 234 benchmarks. Plenty of data behind it.
85Β Β±9
Coding IQ 84.8, 95% interval 76–93. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.12$0.4241K
35.8
Poor value (35.8). Costs more per task than its benchmark results predict. A fair amount of data behind it.
Not measured
136.9
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Qwen3 32B
πŸ‡¨πŸ‡³
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
115.6
From 234 benchmarks. Plenty of data behind it.
85Β Β±9
Coding IQ 84.8, 95% interval 76–93. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.14$0.57131K
35.8
Poor value (35.8). Costs more per task than its benchmark results predict. A fair amount of data behind it.
Not measured
136.9
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Qwen3 32B
πŸ‡¨πŸ‡³
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 Tenstorrent
115.6
From 234 benchmarks. Plenty of data behind it.
85Β Β±9
Coding IQ 84.8, 95% interval 76–93. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.14cache read $0.09$0.4020K
35.8
Poor value (35.8). Costs more per task than its benchmark results predict. A fair amount of data behind it.
Not measured
136.9
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
GPT-4.1 mini
πŸ‡ΊπŸ‡Έ
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 the version names in our catalogue.
115.4
From 375 benchmarks. Plenty of data behind it.
91Β Β±8
Coding IQ 91.3, 95% interval 84–99. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.40cache read $0.10$1.601M
46.2
Poor value (46.2). Costs more per task than its benchmark results predict. Plenty of data behind it.
Not measured
134.2
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
GPT-4.1 mini
πŸ‡ΊπŸ‡Έ
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 the version names in our catalogue.
openai/gpt-4.1-mini:batch Β· batch
115.4
From 375 benchmarks. Plenty of data behind it.
91Β Β±8
Coding IQ 91.3, 95% interval 84–99. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.20cache read $0.05$0.801M
46.2
Poor value (46.2). Costs more per task than its benchmark results predict. Plenty of data behind it.
Not measured
134.2
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
GPT-4.1 mini
πŸ‡ΊπŸ‡Έ
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 the version names in our catalogue.
Inference by Microsoft Azure
115.4
From 375 benchmarks. Plenty of data behind it.
91Β Β±8
Coding IQ 91.3, 95% interval 84–99. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.40cache read $0.10$1.601M
46.2
Poor value (46.2). Costs more per task than its benchmark results predict. Plenty of data behind it.
Not measured
134.2
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
GPT-4.1 mini
πŸ‡ΊπŸ‡Έ
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 the version names in our catalogue.
Inference by OpenAI
115.4
From 375 benchmarks. Plenty of data behind it.
91Β Β±8
Coding IQ 91.3, 95% interval 84–99. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.40cache read $0.10$1.601M
46.2
Poor value (46.2). Costs more per task than its benchmark results predict. Plenty of data behind it.
Not measured
134.2
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
GPT-5 nano
πŸ‡ΊπŸ‡Έ
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Newer version
Newer versionGPT 5.4 NanoWe compared release dates and version names in our catalogue.
115.3
From 332 benchmarks. Plenty of data behind it.
93Β Β±13
Coding IQ 92.7, 95% interval 80–106. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.05cache read $0.005$0.40400K
46.3
Poor value (46.3). Costs more per task than its benchmark results predict. Plenty of data behind it.
51 Β±9
Typical (51 Β±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.
139.1
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
GPT-5 nano
πŸ‡ΊπŸ‡Έ
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Newer version
Newer versionGPT 5.4 NanoWe compared release dates and version names in our catalogue.
openai/gpt-5-nano:batch Β· batch
115.3
From 332 benchmarks. Plenty of data behind it.
93Β Β±13
Coding IQ 92.7, 95% interval 80–106. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.025cache read $0.0025$0.20400K
46.3
Poor value (46.3). Costs more per task than its benchmark results predict. Plenty of data behind it.
51 Β±9
Typical (51 Β±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.
139.1
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
GPT-5 nano
πŸ‡ΊπŸ‡Έ
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Newer version
Newer versionGPT 5.4 NanoWe compared release dates and version names in our catalogue.
Inference by Microsoft Azure
115.3
From 332 benchmarks. Plenty of data behind it.
93Β Β±13
Coding IQ 92.7, 95% interval 80–106. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.05cache read $0.01$0.40400K
46.3
Poor value (46.3). Costs more per task than its benchmark results predict. Plenty of data behind it.
51 Β±9
Typical (51 Β±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.
139.1
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
GPT-5 nano
πŸ‡ΊπŸ‡Έ
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Newer version
Newer versionGPT 5.4 NanoWe compared release dates and version names in our catalogue.
Inference by OpenAI
115.3
From 332 benchmarks. Plenty of data behind it.
93Β Β±13
Coding IQ 92.7, 95% interval 80–106. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.05cache read $0.005$0.40272K
46.3
Poor value (46.3). Costs more per task than its benchmark results predict. Plenty of data behind it.
51 Β±9
Typical (51 Β±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.
139.1
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
GPT-5 nano
πŸ‡ΊπŸ‡Έ
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Newer version
Newer versionGPT 5.4 NanoWe compared release dates and version names in our catalogue.
openai/gpt-5-nano:batch Β· batch
Inference by OpenAI
115.3
From 332 benchmarks. Plenty of data behind it.
93Β Β±13
Coding IQ 92.7, 95% interval 80–106. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.025cache read $0.0025$0.20400K
46.3
Poor value (46.3). Costs more per task than its benchmark results predict. Plenty of data behind it.
51 Β±9
Typical (51 Β±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.
139.1
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
GPT-5 nano
πŸ‡ΊπŸ‡Έ
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Newer version
Newer versionGPT 5.4 NanoWe compared release dates and version names in our catalogue.
flex
Inference by OpenAI
115.3
From 332 benchmarks. Plenty of data behind it.
93Β Β±13
Coding IQ 92.7, 95% interval 80–106. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.025cache read $0.0025$0.20272K
46.3
Poor value (46.3). Costs more per task than its benchmark results predict. Plenty of data behind it.
51 Β±9
Typical (51 Β±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.
139.1
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Mistral Medium 3.5
πŸ‡«πŸ‡·
Maker headquartered in FR, according to the company's own website. The citation is on the model page.
115.3
From 173 benchmarks. Plenty of data behind it.
99Β Β±15
Coding IQ 98.6, 95% interval 84–114. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$1.50$7.50262K
12.0
Poor value (12.0). Costs more per task than its benchmark results predict. Plenty of data behind it.
Not measured
141.6
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Mistral Medium 3.5
πŸ‡«πŸ‡·
Maker headquartered in FR, according to the company's own website. The citation is on the model page.
mistralai/mistral-medium-3-5:batch Β· batch
115.3
From 173 benchmarks. Plenty of data behind it.
99Β Β±15
Coding IQ 98.6, 95% interval 84–114. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.75$3.75262K
12.0
Poor value (12.0). Costs more per task than its benchmark results predict. Plenty of data behind it.
Not measured
141.6
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Mistral Medium 3.5
πŸ‡«πŸ‡·
Maker headquartered in FR, according to the company's own website. The citation is on the model page.
mistralai/mistral-medium-3-5:batch Β· batch
Inference by Mistral AI
115.3
From 173 benchmarks. Plenty of data behind it.
99Β Β±15
Coding IQ 98.6, 95% interval 84–114. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.75$3.75262K
12.0
Poor value (12.0). Costs more per task than its benchmark results predict. Plenty of data behind it.
Not measured
141.6
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Mistral Medium 3.5
πŸ‡«πŸ‡·
Maker headquartered in FR, according to the company's own website. The citation is on the model page.
eu
Inference by Mistral AI
115.3
From 173 benchmarks. Plenty of data behind it.
99Β Β±15
Coding IQ 98.6, 95% interval 84–114. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$1.65$8.25262K
12.0
Poor value (12.0). Costs more per task than its benchmark results predict. Plenty of data behind it.
Not measured
141.6
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Mistral Medium 3.5
πŸ‡«πŸ‡·
Maker headquartered in FR, according to the company's own website. The citation is on the model page.
zdr
Inference by Mistral AI
115.3
From 173 benchmarks. Plenty of data behind it.
99Β Β±15
Coding IQ 98.6, 95% interval 84–114. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$1.50$7.50262K
12.0
Poor value (12.0). Costs more per task than its benchmark results predict. Plenty of data behind it.
Not measured
141.6
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
O1 Mini
πŸ‡ΊπŸ‡Έ
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Newer version
Newer versiono4 MiniWe compared release dates and version names in our catalogue.
115.3
From 187 benchmarks. Plenty of data behind it.
83Β Β±14
Coding IQ 82.8, 95% interval 69–97. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$1.10cache read $0.55$4.40Not published by the provider
32.5
Poor value (32.5). Costs more per task than its benchmark results predict. A fair amount of data behind it.
Not measured
135.3
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
O1 Mini
πŸ‡ΊπŸ‡Έ
Maker headquartered in US, according to the company's own website. The citation is on the model page.
Newer version
Newer versiono4 MiniWe compared release dates and version names in our catalogue.
openai/o1-mini-2024-09-12 Β· 2024-09-12
115.3
From 187 benchmarks. Plenty of data behind it.
83Β Β±14
Coding IQ 82.8, 95% interval 69–97. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$1.10$4.40Not published by the provider
32.5
Poor value (32.5). Costs more per task than its benchmark results predict. A fair amount of data behind it.
Not measured
135.3
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Gemini 2.0 Flash Lite
πŸ‡ΊπŸ‡Έ
Maker headquartered in US, according to the company's own website. The citation is on the model page.
gemini-2.0-flash-lite-001
115.2
From 45 benchmarks. A fair amount of data behind it.
Not enough data yet$0.075$0.30Not published by the providerNot enough data yetNot measuredNot measured
Nova Premier 1.0
πŸ‡ΊπŸ‡Έ
Maker headquartered in US, according to the company's own website. The citation is on the model page.
amazon/nova-premier-v1
β‰ˆ 115
From 13 benchmarks. Little data behind it, so treat it as a rough estimate.
Not enough data yet$2.50cache read $0.625$12.501M
43.5
Poor value (43.5). Costs more per task than its benchmark results predict. Little data behind it, so treat it as a rough estimate.
Not measuredNot measured
Command A
πŸ‡ΊπŸ‡Έ
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.
115.1
From 61 benchmarks. A fair amount of data behind it.
Not enough data yet$2.50$10.00256K
31.8
Poor value (31.8). Costs more per task than its benchmark results predict. Little data behind it, so treat it as a rough estimate.
Not measuredNot measured
Command A
πŸ‡ΊπŸ‡Έ
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 Cohere
115.1
From 61 benchmarks. A fair amount of data behind it.
Not enough data yet$2.50$10.00256K
31.8
Poor value (31.8). Costs more per task than its benchmark results predict. Little data behind it, so treat it as a rough estimate.
Not measuredNot measured
Qwen3 VL 32B Instruct
πŸ‡¨πŸ‡³
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.
115.1
From 73 benchmarks. Plenty of data behind it.
Not enough data yet$0.104$0.416Not published by the providerNot enough data yetNot measuredNot measured
Reka Flash 3
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
115.0
From 12 benchmarks. A fair amount of data behind it.
Not enough data yet$0.10$0.2066K
38.8
Poor value (38.8). Costs more per task than its benchmark results predict. Little data behind it, so treat it as a rough estimate.
Not measuredNot measured
Reka Flash 3
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
Inference by Reka AI
115.0
From 12 benchmarks. A fair amount of data behind it.
Not enough data yet$0.10$0.2066K
38.8
Poor value (38.8). Costs more per task than its benchmark results predict. Little data behind it, so treat it as a rough estimate.
Not measuredNot measured
ERNIE 4.5 300B A47B
πŸ‡¨πŸ‡³
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
114.9
From 12 benchmarks. A fair amount of data behind it.
Not enough data yet$0.28$1.10Not published by the provider
51.4
Fair value (51.4). Costs per task about what its benchmark results predict. Little data behind it, so treat it as a rough estimate.
Not measuredNot measured

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

How this data works