Skip to content

Listings

All listings: Baseten

Every listing for the models on page 1 of 2 of Baseten, including each model's other versions and routes. Back to Baseten

Sold by Baseten

23 listings on this page sold by Baseten.

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
Kimi K3.1 Pro
🇨🇳
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.
142.7
From 161 benchmarks. Plenty of data behind it.
131 ±10
Coding IQ 131.3, 95% interval 121–141. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$2.00cache read $0.20$10.001M
49.2
Fair value (49.2). Costs per task about what its benchmark results predict. Plenty of data behind it.
51 ±9
Typical (51 ±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.
161.2
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
DeepSeek V4.1 Flash 1017
🇨🇳
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.
139.7
From 97 benchmarks. Plenty of data behind it.
127 ±8
Coding IQ 126.8, 95% interval 119–135. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.30cache read $0.007$1.201M
71.4
Good value (71.4). Cheaper per task than its benchmark results predict. Plenty of data behind it.
55 ±9
Typical (55 ±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.
159.4
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
DeepSeek V4.1 Flash 1017
🇨🇳
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.
deepseek-ai/DeepSeek-V4.1-Flash-1017-Fast · fast
139.7
From 97 benchmarks. Plenty of data behind it.
127 ±8
Coding IQ 126.8, 95% interval 119–135. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.60cache read $0.014$2.401M
71.4
Good value (71.4). Cheaper per task than its benchmark results predict. Plenty of data behind it.
55 ±9
Typical (55 ±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.
159.4
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Kimi K3
🇨🇳
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 versionKimi K3.1We compared the version names in our catalogue.
fp8
139.2
From 417 benchmarks. Plenty of data behind it.
124 ±7
Coding IQ 123.8, 95% interval 117–131. 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$15.001M
60.2
Good value (60.2). Cheaper per task than its benchmark results predict. Plenty of data behind it.
55 ±9
Typical (55 ±9). Generates about as many tokens as models with similar coding results. Measured at max effort. Little data behind it, so treat it as a rough estimate.
158.6
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
GLM 5.3
🇨🇳
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.
fp4
138.2
From 318 benchmarks. Plenty of data behind it.
121 ±9
Coding IQ 121.0, 95% interval 112–130. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$1.40cache read $0.14$4.401M
57.7
Good value (57.7). Cheaper per task than its benchmark results predict. Plenty of data behind it.
65 ±10
Chatty (65 ±10). Generates more tokens than models with similar coding results. Measured at max effort. A fair amount of data behind it.
157.1
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
GLM 5.3
🇨🇳
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.
zai-org/GLM-5.3-Fast · fast · fp8
138.2
From 318 benchmarks. Plenty of data behind it.
121 ±9
Coding IQ 121.0, 95% interval 112–130. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$2.10cache read $0.21$6.601M
57.7
Good value (57.7). Cheaper per task than its benchmark results predict. Plenty of data behind it.
65 ±10
Chatty (65 ±10). Generates more tokens than models with similar coding results. Measured at max effort. A fair amount of data behind it.
157.1
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
DeepSeek V4.1 Flash
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
137.7
From 179 benchmarks. Plenty of data behind it.
125 ±16
Coding IQ 124.7, 95% interval 109–140. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.30cache read $0.007$1.201M
68.1
Good value (68.1). Cheaper per task than its benchmark results predict. Plenty of data behind it.
60 ±9
Chatty (60 ±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.
157
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
DeepSeek V4.1 Flash
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
deepseek-ai/DeepSeek-V4.1-Flash-Fast · fast
137.7
From 179 benchmarks. Plenty of data behind it.
125 ±16
Coding IQ 124.7, 95% interval 109–140. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.60cache read $0.014$2.401M
68.1
Good value (68.1). Cheaper per task than its benchmark results predict. Plenty of data behind it.
60 ±9
Chatty (60 ±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.
157
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
DeepSeek V4 Pro 0813
🇨🇳
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
136.8
From 147 benchmarks. Plenty of data behind it.
117 ±7
Coding IQ 117.1, 95% interval 110–124. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$1.32cache read $0.132$3.961M
52.5
Fair value (52.5). Costs per task about what its benchmark results predict. Plenty of data behind it.
58 ±9
Chatty (58 ±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.
156
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Muse Spark Open
🇺🇸
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
136.3
From 58 benchmarks. Plenty of data behind it.
114 ±8
Coding IQ 113.5, 95% interval 105–122. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$1.25cache read $0.15$4.251M
49.6
Fair value (49.6). Costs per task about what its benchmark results predict. Plenty of data behind it.
Not measured
155.5
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
GLM 5.2
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
Newer version
Newer versionGLM 5.3We compared release dates and version names in our catalogue.
fp8
133.9
From 361 benchmarks. Plenty of data behind it.
113 ±9
Coding IQ 113.3, 95% interval 105–122. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$1.40cache read $0.14$4.401M
45.9
Poor value (45.9). Costs more per task than its benchmark results predict. Plenty of data behind it.
58 ±10
Chatty (58 ±10). Generates more tokens than models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
153.1
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
GLM 5.2
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
Newer version
Newer versionGLM 5.3We compared release dates and version names in our catalogue.
zai-org/GLM-5.2-Fast · fast · fp8
133.9
From 361 benchmarks. Plenty of data behind it.
113 ±9
Coding IQ 113.3, 95% interval 105–122. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$2.10cache read $0.21$6.601M
45.9
Poor value (45.9). Costs more per task than its benchmark results predict. Plenty of data behind it.
58 ±10
Chatty (58 ±10). Generates more tokens than models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
153.1
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
GLM 5.3 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.
fp8
133.6
From 266 benchmarks. Plenty of data behind it.
113 ±8
Coding IQ 112.8, 95% interval 105–120. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.15cache read $0.03$0.501M
65.3
Good value (65.3). Cheaper per task than its benchmark results predict. Plenty of data behind it.
68 ±11
Chatty (68 ±11). Generates more tokens than models with similar coding results. Measured at max effort. A fair amount of data behind it.
153.8
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Kimi K2.6
🇨🇳
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 versionKimi K3.1We compared the version names in our catalogue.
132.6
From 305 benchmarks. Plenty of data behind it.
113 ±7
Coding IQ 112.9, 95% interval 106–119. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.95cache read $0.16$4.00262K
50.4
Fair value (50.4). Costs per task about what its benchmark results predict. Plenty of data behind it.
48 ±14
Typical (48 ±14). Generates about as many tokens as models with similar coding results. Measured at default effort. A fair amount of data behind it.
151.6
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
DeepSeek V4 Flash 0731
🇨🇳
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 V4.1 Flash 1017We compared release dates and version names in our catalogue.
fp8
132.2
From 149 benchmarks. Plenty of data behind it.
113 ±8
Coding IQ 112.8, 95% interval 105–121. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.13cache read $0.028$0.261M
64.8
Good value (64.8). Cheaper per task than its benchmark results predict. Plenty of data behind it.
57 ±9
Chatty (57 ±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.
154.4
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
DeepSeek V4 Pro
🇨🇳
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.
130.7
From 303 benchmarks. Plenty of data behind it.
105 ±7
Coding IQ 104.6, 95% interval 98–111. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$1.74cache read $0.145$3.481M
46.9
Poor value (46.9). Costs more per task than its benchmark results predict. Plenty of data behind it.
43 ±10
Terse (43 ±10). Generates fewer tokens than models with similar coding results. Measured at default effort. A fair amount of data behind it.
150.2
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
GLM 5
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
Newer version
Newer versionGLM 5.3We compared release dates and version names in our catalogue.
129.7
From 306 benchmarks. Plenty of data behind it.
107 ±5
Coding IQ 106.7, 95% interval 102–112. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.95$3.15Not published by the provider
49.7
Fair value (49.7). Costs per task about what its benchmark results predict. Plenty of data behind it.
55 ±9
Typical (55 ±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.
146.5
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Kimi K2.5
🇨🇳
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 versionKimi K3.1We compared the version names in our catalogue.
128.5
From 401 benchmarks. Plenty of data behind it.
103 ±5
Coding IQ 102.9, 95% interval 98–108. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.60$3.00Not published by the provider
48.2
Fair value (48.2). Costs per task about what its benchmark results predict. Plenty of data behind it.
42 ±9
Terse (42 ±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.
148.8
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Inkling Small
🇺🇸
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.
127.4
From 169 benchmarks. Plenty of data behind it.
107 ±7
Coding IQ 107.0, 95% interval 101–114. 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$1.201M
45.5
Poor value (45.5). Costs more per task than its benchmark results predict. Plenty of data behind it.
42 ±9
Terse (42 ±9). Generates fewer tokens than models with similar coding results. Measured with reasoning off. Little data behind it, so treat it as a rough estimate.
150.8
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
GLM 4.7
🇨🇳
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 5.3We compared release dates and version names in our catalogue.
127.0
From 234 benchmarks. Plenty of data behind it.
104 ±5
Coding IQ 103.5, 95% interval 98–109. 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$2.20200K
41.0
Poor value (41.0). Costs more per task than its benchmark results predict. Plenty of data behind it.
61 ±9
Chatty (61 ±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.
143.3
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Inkling FP4
🇺🇸
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.
baseten/thinkingmachines/inkling
126.9
From 215 benchmarks. Plenty of data behind it.
102 ±6
Coding IQ 101.9, 95% interval 96–107. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$1.00cache read $0.17$4.051M
27.9
Poor value (27.9). Costs more per task than its benchmark results predict. Plenty of data behind it.
58 ±9
Chatty (58 ±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.
148.4
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Kimi K2.7 Code
🇨🇳
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.
126.9
From 102 benchmarks. Plenty of data behind it.
105 ±9
Coding IQ 104.5, 95% interval 96–113. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.95cache read $0.16$4.00262K
43.6
Poor value (43.6). Costs more per task than its benchmark results predict. Plenty of data behind it.
62 ±9
Chatty (62 ±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.
151.2
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
NVIDIA Nemotron 3.5 Super VL 120B
🇺🇸
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/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B
125.0
From 100 benchmarks. Plenty of data behind it.
104 ±6
Coding IQ 104.2, 95% interval 98–110. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.20cache read $0.04$0.501M
23.2
Poor value (23.2). Costs more per task than its benchmark results predict. Plenty of data behind it.
Not measuredNot measured

Run by Baseten

49 listings on this page where Baseten runs the model. Another provider sells these listings.

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
ModelProviderllminfo IQCoding IQInput $/1MOutput $/1MMax Input TokensValueYapEpoch Capabilities Index
DeepSeek V4.1 Flash 1017
🇨🇳
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.
 Hugging Face
139.7
From 97 benchmarks. Plenty of data behind it.
127 ±8
Coding IQ 126.8, 95% interval 119–135. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.30$1.201M
71.4
Good value (71.4). Cheaper per task than its benchmark results predict. Plenty of data behind it.
55 ±9
Typical (55 ±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.
159.4
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
DeepSeek V4.1 Flash 1017
🇨🇳
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
 NanoGPT
139.7
From 97 benchmarks. Plenty of data behind it.
127 ±8
Coding IQ 126.8, 95% interval 119–135. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.315cache read $0.0074$1.26Not published by the provider
71.4
Good value (71.4). Cheaper per task than its benchmark results predict. Plenty of data behind it.
55 ±9
Typical (55 ±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.
159.4
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
DeepSeek V4.1 Flash 1017
🇨🇳
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.
deepseek/deepseek-v4.1-flash-1017:thinking · thinking · 🇺🇸 US
 NanoGPT
139.7
From 97 benchmarks. Plenty of data behind it.
127 ±8
Coding IQ 126.8, 95% interval 119–135. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.315cache read $0.0074$1.26Not published by the provider
71.4
Good value (71.4). Cheaper per task than its benchmark results predict. Plenty of data behind it.
55 ±9
Typical (55 ±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.
159.4
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
DeepSeek V4.1 Flash 1017
🇨🇳
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.
 OpenRouter
139.7
From 97 benchmarks. Plenty of data behind it.
127 ±8
Coding IQ 126.8, 95% interval 119–135. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.30cache read $0.007$1.201M
71.4
Good value (71.4). Cheaper per task than its benchmark results predict. Plenty of data behind it.
55 ±9
Typical (55 ±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.
159.4
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
DeepSeek V4.1 Flash 1017
🇨🇳
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.
fast
 OpenRouter
139.7
From 97 benchmarks. Plenty of data behind it.
127 ±8
Coding IQ 126.8, 95% interval 119–135. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.60cache read $0.014$2.401M
71.4
Good value (71.4). Cheaper per task than its benchmark results predict. Plenty of data behind it.
55 ±9
Typical (55 ±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.
159.4
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Kimi K3
🇨🇳
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 versionKimi K3.1We compared the version names in our catalogue.
 Hugging Face
139.2
From 417 benchmarks. Plenty of data behind it.
124 ±7
Coding IQ 123.8, 95% interval 117–131. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$3.00$15.001M
60.2
Good value (60.2). Cheaper per task than its benchmark results predict. Plenty of data behind it.
55 ±9
Typical (55 ±9). Generates about as many tokens as models with similar coding results. Measured at max effort. Little data behind it, so treat it as a rough estimate.
158.6
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Kimi K3
🇨🇳
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 versionKimi K3.1We compared the version names in our catalogue.
fp8 · 🇺🇸 US
 NanoGPT
139.2
From 417 benchmarks. Plenty of data behind it.
124 ±7
Coding IQ 123.8, 95% interval 117–131. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$3.15cache read $0.315$15.75Not published by the provider
60.2
Good value (60.2). Cheaper per task than its benchmark results predict. Plenty of data behind it.
55 ±9
Typical (55 ±9). Generates about as many tokens as models with similar coding results. Measured at max effort. Little data behind it, so treat it as a rough estimate.
158.6
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Kimi K3
🇨🇳
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 versionKimi K3.1We compared the version names in our catalogue.
fp8
 OpenRouter
139.2
From 417 benchmarks. Plenty of data behind it.
124 ±7
Coding IQ 123.8, 95% interval 117–131. 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$15.001M
60.2
Good value (60.2). Cheaper per task than its benchmark results predict. Plenty of data behind it.
55 ±9
Typical (55 ±9). Generates about as many tokens as models with similar coding results. Measured at max effort. Little data behind it, so treat it as a rough estimate.
158.6
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
GLM 5.3
🇨🇳
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.
 Hugging Face
138.2
From 318 benchmarks. Plenty of data behind it.
121 ±9
Coding IQ 121.0, 95% interval 112–130. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$1.40$4.401M
57.7
Good value (57.7). Cheaper per task than its benchmark results predict. Plenty of data behind it.
65 ±10
Chatty (65 ±10). Generates more tokens than models with similar coding results. Measured at max effort. A fair amount of data behind it.
157.1
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
GLM 5.3
🇨🇳
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.
fp4 · 🇺🇸 US
 NanoGPT
138.2
From 318 benchmarks. Plenty of data behind it.
121 ±9
Coding IQ 121.0, 95% interval 112–130. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$1.47cache read $0.147$4.62Not published by the provider
57.7
Good value (57.7). Cheaper per task than its benchmark results predict. Plenty of data behind it.
65 ±10
Chatty (65 ±10). Generates more tokens than models with similar coding results. Measured at max effort. A fair amount of data behind it.
157.1
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
GLM 5.3
🇨🇳
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.
z-ai/glm-5.3:thinking · thinking · fp4 · 🇺🇸 US
 NanoGPT
138.2
From 318 benchmarks. Plenty of data behind it.
121 ±9
Coding IQ 121.0, 95% interval 112–130. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$1.47cache read $0.147$4.62Not published by the provider
57.7
Good value (57.7). Cheaper per task than its benchmark results predict. Plenty of data behind it.
65 ±10
Chatty (65 ±10). Generates more tokens than models with similar coding results. Measured at max effort. A fair amount of data behind it.
157.1
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
GLM 5.3
🇨🇳
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.
fast · fp8
 OpenRouter
138.2
From 318 benchmarks. Plenty of data behind it.
121 ±9
Coding IQ 121.0, 95% interval 112–130. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$2.10cache read $0.21$6.601M
57.7
Good value (57.7). Cheaper per task than its benchmark results predict. Plenty of data behind it.
65 ±10
Chatty (65 ±10). Generates more tokens than models with similar coding results. Measured at max effort. A fair amount of data behind it.
157.1
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
GLM 5.3
🇨🇳
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.
fp4
 OpenRouter
138.2
From 318 benchmarks. Plenty of data behind it.
121 ±9
Coding IQ 121.0, 95% interval 112–130. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$1.40cache read $0.14$4.401M
57.7
Good value (57.7). Cheaper per task than its benchmark results predict. Plenty of data behind it.
65 ±10
Chatty (65 ±10). Generates more tokens than models with similar coding results. Measured at max effort. A fair amount of data behind it.
157.1
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
GLM 5.3
🇨🇳
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.
fp4/us · fp4
 OpenRouter
138.2
From 318 benchmarks. Plenty of data behind it.
121 ±9
Coding IQ 121.0, 95% interval 112–130. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$1.40cache read $0.14$4.401M
57.7
Good value (57.7). Cheaper per task than its benchmark results predict. Plenty of data behind it.
65 ±10
Chatty (65 ±10). Generates more tokens than models with similar coding results. Measured at max effort. A fair amount of data behind it.
157.1
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
GLM 5.3
🇨🇳
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/fast · fp8
 OpenRouter
138.2
From 318 benchmarks. Plenty of data behind it.
121 ±9
Coding IQ 121.0, 95% interval 112–130. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$2.10cache read $0.21$6.601M
57.7
Good value (57.7). Cheaper per task than its benchmark results predict. Plenty of data behind it.
65 ±10
Chatty (65 ±10). Generates more tokens than models with similar coding results. Measured at max effort. A fair amount of data behind it.
157.1
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
DeepSeek V4.1 Flash
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
 Hugging Face
137.7
From 179 benchmarks. Plenty of data behind it.
125 ±16
Coding IQ 124.7, 95% interval 109–140. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.30$1.201M
68.1
Good value (68.1). Cheaper per task than its benchmark results predict. Plenty of data behind it.
60 ±9
Chatty (60 ±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.
157
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
DeepSeek V4.1 Flash
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
🇺🇸 US
 NanoGPT
137.7
From 179 benchmarks. Plenty of data behind it.
125 ±16
Coding IQ 124.7, 95% interval 109–140. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.315cache read $0.0074$1.26Not published by the provider
68.1
Good value (68.1). Cheaper per task than its benchmark results predict. Plenty of data behind it.
60 ±9
Chatty (60 ±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.
157
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
DeepSeek V4.1 Flash
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
deepseek/deepseek-v4.1-flash:thinking · thinking · 🇺🇸 US
 NanoGPT
137.7
From 179 benchmarks. Plenty of data behind it.
125 ±16
Coding IQ 124.7, 95% interval 109–140. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.315cache read $0.0074$1.26Not published by the provider
68.1
Good value (68.1). Cheaper per task than its benchmark results predict. Plenty of data behind it.
60 ±9
Chatty (60 ±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.
157
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
DeepSeek V4 Pro 0813
🇨🇳
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.
 Hugging Face
136.8
From 147 benchmarks. Plenty of data behind it.
117 ±7
Coding IQ 117.1, 95% interval 110–124. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$1.32$3.961M
52.5
Fair value (52.5). Costs per task about what its benchmark results predict. Plenty of data behind it.
58 ±9
Chatty (58 ±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.
156
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
DeepSeek V4 Pro 0813
🇨🇳
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 · 🇺🇸 US
 NanoGPT
136.8
From 147 benchmarks. Plenty of data behind it.
117 ±7
Coding IQ 117.1, 95% interval 110–124. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$1.386cache read $0.1386$4.158Not published by the provider
52.5
Fair value (52.5). Costs per task about what its benchmark results predict. Plenty of data behind it.
58 ±9
Chatty (58 ±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.
156
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
DeepSeek V4 Pro 0813
🇨🇳
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.
deepseek/deepseek-v4-pro-0813:thinking · thinking · fp4 · 🇺🇸 US
 NanoGPT
136.8
From 147 benchmarks. Plenty of data behind it.
117 ±7
Coding IQ 117.1, 95% interval 110–124. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$1.386cache read $0.1386$4.158Not published by the provider
52.5
Fair value (52.5). Costs per task about what its benchmark results predict. Plenty of data behind it.
58 ±9
Chatty (58 ±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.
156
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
DeepSeek V4 Pro 0813
🇨🇳
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
 OpenRouter
136.8
From 147 benchmarks. Plenty of data behind it.
117 ±7
Coding IQ 117.1, 95% interval 110–124. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$1.32cache read $0.132$3.961M
52.5
Fair value (52.5). Costs per task about what its benchmark results predict. Plenty of data behind it.
58 ±9
Chatty (58 ±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.
156
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Muse Spark Open
🇺🇸
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 · 🇺🇸 US
 NanoGPT
136.3
From 58 benchmarks. Plenty of data behind it.
114 ±8
Coding IQ 113.5, 95% interval 105–122. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$1.3125cache read $0.1575$4.4625Not published by the provider
49.6
Fair value (49.6). Costs per task about what its benchmark results predict. Plenty of data behind it.
Not measured
155.5
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Muse Spark Open
🇺🇸
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
 OpenRouter
136.3
From 58 benchmarks. Plenty of data behind it.
114 ±8
Coding IQ 113.5, 95% interval 105–122. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$1.25cache read $0.15$4.251M
49.6
Fair value (49.6). Costs per task about what its benchmark results predict. Plenty of data behind it.
Not measured
155.5
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
Muse Spark Open 816B A42B
🇺🇸
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.
 Hugging Face
134.2
From 13 benchmarks. Plenty of data behind it.
≈ 113 ±10
Coding IQ 113.1, 95% interval 103–124. 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.
$1.25$4.251M
56.0
Good value (56.0). Cheaper per task than its benchmark results predict. Little data behind it, so treat it as a rough estimate.
54 ±9
Typical (54 ±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
GLM 5.2
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
Newer version
Newer versionGLM 5.3We compared release dates and version names in our catalogue.
 Hugging Face
133.9
From 361 benchmarks. Plenty of data behind it.
113 ±9
Coding IQ 113.3, 95% interval 105–122. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$1.40$4.401M
45.9
Poor value (45.9). Costs more per task than its benchmark results predict. Plenty of data behind it.
58 ±10
Chatty (58 ±10). Generates more tokens than models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
153.1
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
GLM 5.2
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
Newer version
Newer versionGLM 5.3We compared release dates and version names in our catalogue.
fp8 · 🇺🇸 US
 NanoGPT
133.9
From 361 benchmarks. Plenty of data behind it.
113 ±9
Coding IQ 113.3, 95% interval 105–122. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$1.47cache read $0.147$4.62Not published by the provider
45.9
Poor value (45.9). Costs more per task than its benchmark results predict. Plenty of data behind it.
58 ±10
Chatty (58 ±10). Generates more tokens than models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
153.1
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
GLM 5.2
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
Newer version
Newer versionGLM 5.3We compared release dates and version names in our catalogue.
z-ai/glm-5.2:thinking · thinking · fp8 · 🇺🇸 US
 NanoGPT
133.9
From 361 benchmarks. Plenty of data behind it.
113 ±9
Coding IQ 113.3, 95% interval 105–122. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$1.47cache read $0.147$4.62Not published by the provider
45.9
Poor value (45.9). Costs more per task than its benchmark results predict. Plenty of data behind it.
58 ±10
Chatty (58 ±10). Generates more tokens than models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
153.1
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
GLM 5.2
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
Newer version
Newer versionGLM 5.3We compared release dates and version names in our catalogue.
fast · fp8
 OpenRouter
133.9
From 361 benchmarks. Plenty of data behind it.
113 ±9
Coding IQ 113.3, 95% interval 105–122. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$2.10cache read $0.21$6.601M
45.9
Poor value (45.9). Costs more per task than its benchmark results predict. Plenty of data behind it.
58 ±10
Chatty (58 ±10). Generates more tokens than models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
153.1
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
GLM 5.2
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
Newer version
Newer versionGLM 5.3We compared release dates and version names in our catalogue.
fp8
 OpenRouter
133.9
From 361 benchmarks. Plenty of data behind it.
113 ±9
Coding IQ 113.3, 95% interval 105–122. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$1.40cache read $0.14$4.401M
45.9
Poor value (45.9). Costs more per task than its benchmark results predict. Plenty of data behind it.
58 ±10
Chatty (58 ±10). Generates more tokens than models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
153.1
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
GLM 5.2
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
Newer version
Newer versionGLM 5.3We compared release dates and version names in our catalogue.
fp8/us · fp8
 OpenRouter
133.9
From 361 benchmarks. Plenty of data behind it.
113 ±9
Coding IQ 113.3, 95% interval 105–122. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$1.40cache read $0.14$4.401M
45.9
Poor value (45.9). Costs more per task than its benchmark results predict. Plenty of data behind it.
58 ±10
Chatty (58 ±10). Generates more tokens than models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
153.1
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
GLM 5.2
🇨🇳
Maker headquartered in CN, according to the company's own website. The citation is on the model page.
Newer version
Newer versionGLM 5.3We compared release dates and version names in our catalogue.
us/fast · fp8
 OpenRouter
133.9
From 361 benchmarks. Plenty of data behind it.
113 ±9
Coding IQ 113.3, 95% interval 105–122. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$2.10cache read $0.21$6.601M
45.9
Poor value (45.9). Costs more per task than its benchmark results predict. Plenty of data behind it.
58 ±10
Chatty (58 ±10). Generates more tokens than models with similar coding results. Measured at default effort. Little data behind it, so treat it as a rough estimate.
153.1
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
GLM 5.3 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.
 Hugging Face
133.6
From 266 benchmarks. Plenty of data behind it.
113 ±8
Coding IQ 112.8, 95% interval 105–120. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.15$0.501M
65.3
Good value (65.3). Cheaper per task than its benchmark results predict. Plenty of data behind it.
68 ±11
Chatty (68 ±11). Generates more tokens than models with similar coding results. Measured at max effort. A fair amount of data behind it.
153.8
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
GLM 5.3 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.
fp8 · 🇺🇸 US
 NanoGPT
133.6
From 266 benchmarks. Plenty of data behind it.
113 ±8
Coding IQ 112.8, 95% interval 105–120. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.1575cache read $0.0315$0.525Not published by the provider
65.3
Good value (65.3). Cheaper per task than its benchmark results predict. Plenty of data behind it.
68 ±11
Chatty (68 ±11). Generates more tokens than models with similar coding results. Measured at max effort. A fair amount of data behind it.
153.8
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
GLM 5.3 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.
fp8
 OpenRouter
133.6
From 266 benchmarks. Plenty of data behind it.
113 ±8
Coding IQ 112.8, 95% interval 105–120. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.15cache read $0.03$0.501M
65.3
Good value (65.3). Cheaper per task than its benchmark results predict. Plenty of data behind it.
68 ±11
Chatty (68 ±11). Generates more tokens than models with similar coding results. Measured at max effort. A fair amount of data behind it.
153.8
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
DeepSeek V4 Flash 0731
🇨🇳
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 V4.1 Flash 1017We compared release dates and version names in our catalogue.
 Hugging Face
132.2
From 149 benchmarks. Plenty of data behind it.
113 ±8
Coding IQ 112.8, 95% interval 105–121. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.13$0.261M
64.8
Good value (64.8). Cheaper per task than its benchmark results predict. Plenty of data behind it.
57 ±9
Chatty (57 ±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.
154.4
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
DeepSeek V4 Flash 0731
🇨🇳
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 V4.1 Flash 1017We compared release dates and version names in our catalogue.
fp8 · 🇺🇸 US
 NanoGPT
132.2
From 149 benchmarks. Plenty of data behind it.
113 ±8
Coding IQ 112.8, 95% interval 105–121. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.1365cache read $0.0294$0.273Not published by the provider
64.8
Good value (64.8). Cheaper per task than its benchmark results predict. Plenty of data behind it.
57 ±9
Chatty (57 ±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.
154.4
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
DeepSeek V4 Flash 0731
🇨🇳
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 V4.1 Flash 1017We compared release dates and version names in our catalogue.
deepseek/deepseek-v4-flash-0731:thinking · thinking · fp8 · 🇺🇸 US
 NanoGPT
132.2
From 149 benchmarks. Plenty of data behind it.
113 ±8
Coding IQ 112.8, 95% interval 105–121. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.1365cache read $0.0294$0.273Not published by the provider
64.8
Good value (64.8). Cheaper per task than its benchmark results predict. Plenty of data behind it.
57 ±9
Chatty (57 ±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.
154.4
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
DeepSeek V4 Flash 0731
🇨🇳
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 V4.1 Flash 1017We compared release dates and version names in our catalogue.
fp32
 OpenRouter
132.2
From 149 benchmarks. Plenty of data behind it.
113 ±8
Coding IQ 112.8, 95% interval 105–121. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.13cache read $0.028$0.261M
64.8
Good value (64.8). Cheaper per task than its benchmark results predict. Plenty of data behind it.
57 ±9
Chatty (57 ±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.
154.4
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
DeepSeek V4 Flash 0731
🇨🇳
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 V4.1 Flash 1017We compared release dates and version names in our catalogue.
fp8
 OpenRouter
132.2
From 149 benchmarks. Plenty of data behind it.
113 ±8
Coding IQ 112.8, 95% interval 105–121. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.13cache read $0.028$0.261M
64.8
Good value (64.8). Cheaper per task than its benchmark results predict. Plenty of data behind it.
57 ±9
Chatty (57 ±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.
154.4
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
DeepSeek V4.1 Flash 0910
🇨🇳
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.
Later snapshot
Later snapshotDeepSeek V4.1 Flash 1017We compared the dates in the model names.
deepseek/deepseek-v4.1-flash
 OpenRouter
≈ 131
From 67 benchmarks. Little data behind it, so treat it as a rough estimate.
Not enough data yet$0.30cache read $0.007$1.201MNot enough data yetNot measuredNot measured
DeepSeek V4.1 Flash 0910
🇨🇳
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.
Later snapshot
Later snapshotDeepSeek V4.1 Flash 1017We compared the dates in the model names.
deepseek/deepseek-v4.1-flash · us
 OpenRouter
≈ 131
From 67 benchmarks. Little data behind it, so treat it as a rough estimate.
Not enough data yet$0.30cache read $0.007$1.201MNot enough data yetNot measuredNot measured
Beam 1.0 501B A23B
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
 Hugging Face
130.0
From 84 benchmarks. Plenty of data behind it.
109 ±6
Coding IQ 109.0, 95% interval 103–115. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.40$1.201MNot enough data yetNot measuredNot measured
Beam 1.0 501B A23B
Open weights. The weights are publicly available, according to the provider's model list or a public Hugging Face repo.
fp4
 OpenRouter
130.0
From 84 benchmarks. Plenty of data behind it.
109 ±6
Coding IQ 109.0, 95% interval 103–115. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.40cache read $0.04$1.201MNot enough data yetNot measuredNot measured
DeepSeek V4 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 versionDeepSeek V4.1 Flash 1017We compared release dates and version names in our catalogue.
deepseek/deepseek-v4-flash-latest · latest · 🇺🇸 US
 NanoGPT
129.1
From 229 benchmarks. Plenty of data behind it.
108 ±12
Coding IQ 107.8, 95% interval 95–120. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.315cache read $0.0074$1.26Not published by the provider
74.6
Good value (74.6). Cheaper per task than its benchmark results predict. Plenty of data behind it.
44 ±9
Terse (44 ±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.
146.9
Epoch Capabilities IndexSource: Epoch AI Benchmarking Hub API
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 · 🇺🇸 US
 NanoGPT
125.9
From 18 benchmarks. A fair amount of data behind it.
Not enough data yet$0.21cache read $0.042$0.525Not published by the provider
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
 OpenRouter
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
NVIDIA Nemotron 3.5 Super VL 120B
🇺🇸
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/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16 · bf16
 Hugging Face
125.0
From 100 benchmarks. Plenty of data behind it.
104 ±6
Coding IQ 104.2, 95% interval 98–110. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.20$0.501M
23.2
Poor value (23.2). Costs more per task than its benchmark results predict. Plenty of data behind it.
Not measuredNot measured
NVIDIA Nemotron 3.5 Super VL 120B
🇺🇸
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/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-NVFP4 · nvfp4
 Hugging Face
125.0
From 100 benchmarks. Plenty of data behind it.
104 ±6
Coding IQ 104.2, 95% interval 98–110. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.
$0.20$0.501M
23.2
Poor value (23.2). Costs more per task than its benchmark results predict. Plenty of data behind it.
Not measuredNot measured

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

How this data works