Epoch Capabilities Index
One number per model, fit by Epoch over the benchmarks that model has results on, with the interval around it. 258 models placed, 116 of them listed in the catalog Vitna prices.
retrieved ·CC-BY-4.0epoch.ai/eci ↗
What the interval says
TOP 24 OF 258Claude Fable 5.2 leads at 173.4, and 8 other models have intervals reaching its own, so the index does not separate them.
- 01Claude Fable 5.2173.4
- 02GPT-6.2 Astra172.8
- 03Gemini 4 Carbon169.7
- 04GPT-6 Astra169.2
- 05Claude Opus 5.5168.7
- 06GPT-6.1 Sol168.6
- 07GPT-6.1 Terra167.7
- 08Claude Fable 5.1167.1
- 09Gemini 4 Hydrogen166.8
- 10Claude Sonnet 5.5166.6
- 11Gemini 4 Argon166.5
- 12Claude Opus 5164.7
- 13Claude Fable 5164.2
- 14GPT-6 Sol163.8
- 15GPT-5.6 Sol163.0
- 16GPT-5.5 Pro162.8
- 17Muse Wonder161.7
- 18Kimi K3.1 Pro161.2
- 19GPT-5.6 Terra160.8
- 20Claude Haiku 5.5160.3
- 21GPT-5.4 Pro160.3
- 22GPT-5.5160.2
- 23Claude Opus 4.8159.9
- 24DeepSeek V4.1 Flash 1017159.4
The rail runs 157.1 to 177.2, the range these 24 rows cover rather than the whole index, because scaled to all 258 they would sit on top of one another. Each bar is the interval Epoch publishes and the mark is the figure; a marked row is one whose interval reaches that of the leader. Epoch states that absolute values are meaningless by themselves and that comparisons between models are what the index is for, that values are not linearly related to benchmark accuracy, and that a highly specialised model may score low while being capable in its own domain.
Vitna fits a scale of its own over the six openly licensed boards, 319 ranked runs over 46 columns, and this is the published index the ordering of that fit is checked against. The two are separate fits over different tables, so their numbers do not compare and only their orderings do.
All 258, highest first.
Index figures, as Epoch publishes them| # | Model | Organisation | Weights | Released | Results | Interval | Index |
|---|---|---|---|---|---|---|---|
| 1 | Claude Fable 5.2 | Anthropic | Closed weights | 21 | 170.0 to 177.2 | 173.4 | |
| 2 | GPT-6.2 Astra | OpenAI | Closed weights | 15 | 169.4 to 176.9 | 172.8 | |
| 3 | Gemini 4 Carbon | Google DeepMind | Closed weights | 17 | 166.6 to 173.1 | 169.7 | |
| 4 | GPT-6 Astra | OpenAI | Closed weights | 26 | 166.4 to 172.5 | 169.2 | |
| 5 | Claude Opus 5.5 | Anthropic | Closed weights | 23 | 165.9 to 172.0 | 168.7 | |
| 6 | GPT-6.1 Sol | OpenAI | Closed weights | 21 | 165.7 to 171.9 | 168.6 | |
| 7 | GPT-6.1 Terra | OpenAI | Closed weights | 13 | 164.9 to 171.0 | 167.7 | |
| 8 | Claude Fable 5.1 | Anthropic | Closed weights | 25 | 164.4 to 170.1 | 167.1 | |
| 9 | Gemini 4 Hydrogen | Google DeepMind | Closed weights | 13 | 164.0 to 170.1 | 166.8 | |
| 10 | Claude Sonnet 5.5 | Anthropic | Closed weights | 20 | 163.9 to 169.6 | 166.6 | |
| 11 | Gemini 4 Argon | Google DeepMind | Closed weights | 22 | 163.9 to 169.5 | 166.5 | |
| 12 | Claude Opus 5 | Anthropic | Closed weights | 24 | 162.3 to 167.5 | 164.7 | |
| 13 | Claude Fable 5 | Anthropic | Closed weights | 26 | 161.8 to 166.9 | 164.2 | |
| 14 | GPT-6 Sol | OpenAI | Closed weights | 20 | 161.4 to 166.6 | 163.8 | |
| 15 | GPT-5.6 Sol | OpenAI | Closed weights | 27 | 160.7 to 165.6 | 163.0 | |
| 16 | GPT-5.5 Pro | OpenAI | Closed weights | 10 | 160.1 to 165.9 | 162.8 | |
| 17 | Muse Wonder | Meta AI | Closed weights | 13 | 159.4 to 164.3 | 161.7 | |
| 18 | Kimi K3.1 Pro | Moonshot | Open weights | 16 | 158.9 to 163.7 | 161.2 | |
| 19 | GPT-5.6 Terra | OpenAI | Closed weights | 20 | 158.5 to 163.3 | 160.8 | |
| 20 | Claude Haiku 5.5 | Anthropic | Closed weights | 20 | 158.2 to 162.7 | 160.3 | |
| 21 | GPT-5.4 Pro | OpenAI | Closed weights | 9 | 157.9 to 163.0 | 160.3 | |
| 22 | GPT-5.5 | OpenAI | Closed weights | 28 | 158.2 to 162.4 | 160.2 | |
| 23 | Claude Opus 4.8 | Anthropic | Closed weights | 26 | 157.8 to 162.2 | 159.9 | |
| 24 | DeepSeek V4.1 Flash 1017 | DeepSeek | Open weights | 11 | 157.1 to 161.9 | 159.4 | |
| 25 | Gemini 3.8 Flash | Google DeepMind | Closed weights | 23 | 156.8 to 161.0 | 158.8 | |
| 26 | Kimi K3 | Moonshot | Open weights | 22 | 156.7 to 160.8 | 158.6 | |
| 27 | Gemini 3.7 Flash | Google DeepMind | Closed weights | 18 | 156.5 to 160.6 | 158.4 | |
| 28 | Muse Spark 1.3 | Meta AI | Closed weights | 14 | 156.2 to 160.5 | 158.2 | |
| 29 | GPT-5.4 | OpenAI | Closed weights | 24 | 156.0 to 160.0 | 157.9 | |
| 30 | Grok 4.6 | xAI | Closed weights | 20 | 155.9 to 160.0 | 157.8 | |
| 31 | Qwen 3.8 Max | Alibaba | Closed weights | 13 | 155.5 to 159.8 | 157.6 | |
| 32 | Claude Opus 4.7 | Anthropic | Closed weights | 24 | 155.6 to 159.6 | 157.5 | |
| 33 | GPT-5.2 Pro | OpenAI | Closed weights | 6 | 155.1 to 159.9 | 157.3 | |
| 34 | GLM-5.3 | Z.ai (Zhipu AI) | Open weights | 16 | 155.0 to 159.4 | 157.1 | |
| 35 | DeepSeek V4.1 Flash | DeepSeek | Open weights | 20 | 155.1 to 159.2 | 157.0 | |
| 36 | Claude Sonnet 5 | Anthropic | Closed weights | 18 | 154.9 to 159.1 | 156.9 | |
| 37 | GPT-5.6 Luna | OpenAI | Closed weights | 21 | 154.9 to 158.8 | 156.7 | |
| 38 | GPT-6 Luna | OpenAI | Closed weights | 18 | 154.6 to 158.5 | 156.4 | |
| 39 | Grok 4.7 | xAI | Closed weights | 19 | 154.5 to 158.5 | 156.4 | |
| 40 | Claude Opus 4.6 | Anthropic | Closed weights | 27 | 154.6 to 158.3 | 156.3 | |
| 41 | Gemini 3.1 Pro | Google DeepMind | Closed weights | 28 | 154.5 to 158.2 | 156.3 | |
| 42 | Qwen3.8 Max (0902) | Alibaba | Closed weights | 9 | 153.9 to 158.3 | 156.0 | |
| 43 | DeepSeek V4 Pro 0813 | DeepSeek | Open weights | 16 | 154.1 to 158.1 | 156.0 | |
| 44 | Muse Spark Open | Meta AI | Open weights | 8 | 153.3 to 157.8 | 155.5 | |
| 45 | Gemini 3.5 Flash | Google DeepMind | Closed weights | 18 | 153.5 to 157.3 | 155.3 | |
| 46 | Gemini 3.6 Flash | Google DeepMind | Closed weights | 17 | 153.4 to 157.3 | 155.2 | |
| 47 | Muse Spark 1.2 | Meta AI | Closed weights | 9 | 152.7 to 157.5 | 155.0 | |
| 48 | Grok 4.5 | xAI | Closed weights | 19 | 153.1 to 156.9 | 155.0 | |
| 49 | Muse Spark 1.1 | Meta AI | Closed weights | 7 | 152.4 to 157.7 | 154.9 | |
| 50 | Mistral Large 4 | Mistral AI | Open weights | 7 | 151.8 to 157.4 | 154.6 | |
| 50 | Qwen3.7-Max | Alibaba | Closed weights | 12 | 152.6 to 156.8 | 154.6 | |
| 52 | GPT-5.2 | OpenAI | Closed weights | 26 | 152.9 to 156.3 | 154.5 | |
| 53 | DeepSeek V4 Flash 0731 | DeepSeek | Open weights | 15 | 152.6 to 156.4 | 154.4 | |
| 54 | GLM-5.3-Flash | Z.ai (Zhipu AI) | Open weights | 14 | 151.9 to 155.8 | 153.8 | |
| 55 | Muse Spark | Meta AI | Closed weights | 5 | 150.9 to 155.5 | 153.2 | |
| 56 | GLM-5.2 | Z.ai (Zhipu AI) | Open weights | 20 | 151.3 to 154.9 | 153.1 | |
| 57 | Gemini 3 Pro | Google DeepMind | Closed weights | 19 | 151.5 to 154.8 | 153.0 | |
| 58 | Claude Sonnet 4.6 | Anthropic | Closed weights | 21 | 151.2 to 154.5 | 152.8 | |
| 59 | Mistral Large 4 Preview | Mistral AI | Open weights | 16 | 150.8 to 154.7 | 152.7 | |
| 60 | Qwen 3.8 27B | Alibaba | Open weights | 9 | 150.7 to 154.7 | 152.6 | |
| 61 | Gemini 3 Flash | Google DeepMind | Closed weights | 20 | 150.8 to 154.1 | 152.3 | |
| 62 | Grok 4.20 | xAI | Closed weights | 15 | 150.5 to 154.1 | 152.3 | |
| 63 | Claude Opus 4.5 | Anthropic | Closed weights | 29 | 150.3 to 153.4 | 151.9 | |
| 64 | Kimi K2.6 | Moonshot | Open weights | 15 | 149.9 to 153.4 | 151.6 | |
| 65 | GPT-5 Pro | OpenAI | Closed weights | 7 | 149.4 to 153.3 | 151.3 | |
| 66 | Kimi K2.7 Code | Moonshot | Open weights | 11 | 149.2 to 153.3 | 151.2 | |
| 67 | GPT-5.1 | OpenAI | Closed weights | 19 | 149.3 to 152.6 | 150.9 | |
| 68 | Beam 1.0 | Reflection | Open weights | 4 | 148.6 to 153.1 | 150.8 | |
| 69 | Inkling-Small | Thinking Machines | Open weights | 9 | 149.0 to 152.5 | 150.8 | |
| 70 | GLM-5.1 | Z.ai (Zhipu AI) | Open weights | 11 | 148.7 to 152.8 | 150.7 | |
| 71 | DeepSeek-V4-Pro | DeepSeek | Open weights | 14 | 148.4 to 152.0 | 150.2 | |
| 72 | GPT-5 | OpenAI | Closed weights | 28 | anchor, fixed | 150.0 | |
| 73 | Grok 4.3 Beta | xAI | Closed weights | 11 | 148.0 to 151.9 | 149.9 | |
| 74 | Qwen 3.6 Max (Preview) | Alibaba | Other | 9 | 147.9 to 151.9 | 149.9 | |
| 75 | GPT-5.4 Mini | OpenAI | Closed weights | 15 | 147.7 to 151.1 | 149.4 | |
| 76 | Qwen 3.6 Plus | Alibaba | Closed weights | 10 | 146.8 to 150.9 | 148.9 | |
| 77 | Kimi K2.5 | Moonshot | Open weights | 14 | 147.3 to 150.3 | 148.8 | |
| 78 | Inkling | Thinking Machines | Open weights | 17 | 146.8 to 150.1 | 148.4 | |
| 79 | Qwen3.5 397B-A17B | Alibaba | Open weights | 7 | 145.9 to 150.1 | 148.0 | |
| 80 | Qwen3.7-Plus | Alibaba | Closed weights | 8 | 146.0 to 150.0 | 147.9 | |
| 81 | MiniMax-M3 | MiniMax | Open weights | 12 | 145.7 to 150.0 | 147.8 | |
| 82 | Claude Sonnet 4.5 | Anthropic | Closed weights | 26 | 146.1 to 149.1 | 147.6 | |
| 83 | o3-pro | OpenAI | Closed weights | 7 | 145.7 to 149.4 | 147.5 | |
| 84 | Qwen 3.5 Plus (hosted 397B-A17B) | Alibaba | Other | 11 | 145.4 to 149.2 | 147.3 | |
| 85 | Nemotron 3 Ultra | Nvidia | Open weights | 10 | 144.9 to 149.2 | 147.0 | |
| 85 | o3 | OpenAI | Closed weights | 25 | 145.5 to 148.6 | 147.0 | |
| 87 | DeepSeek-V4-Flash | DeepSeek | Open weights | 4 | 143.7 to 150.2 | 146.9 | |
| 87 | Grok 4 | xAI | Closed weights | 17 | 145.4 to 148.4 | 146.9 | |
| 89 | Qwen3.6 27B | Alibaba | Open weights | 6 | 144.6 to 148.9 | 146.7 | |
| 90 | GLM-5 | Z.ai (Zhipu AI) | Open weights | 10 | 144.9 to 148.1 | 146.5 | |
| 91 | GPT-5.4 Nano | OpenAI | Closed weights | 13 | 144.7 to 148.0 | 146.4 | |
| 92 | DeepSeek-V3.2 | DeepSeek | Open weights | 14 | 144.7 to 147.8 | 146.3 | |
| 93 | GPT-5 mini | OpenAI | Closed weights | 17 | 144.5 to 147.6 | 146.0 | |
| 94 | Gemini 3.5 Flash-Lite | Google DeepMind | Closed weights | 13 | 144.0 to 147.2 | 145.6 | |
| 95 | Kimi K2 Thinking | Moonshot | Open weights | 9 | 143.7 to 147.3 | 145.5 | |
| 96 | o4-mini | OpenAI | Closed weights | 22 | 143.7 to 146.7 | 145.2 | |
| 97 | Gemini 3.1 Flash-Lite | Closed weights | 8 | 142.8 to 147.1 | 145.0 | ||
| 98 | Gemini 2.5 Pro (Jun 2025) | Google DeepMind | Closed weights | 20 | 143.5 to 146.3 | 144.9 | |
| 99 | Claude Opus 4.1 | Anthropic | Closed weights | 16 | 143.0 to 146.5 | 144.8 | |
| 100 | DeepSeek-V3.2-Exp | DeepSeek | Open weights | 8 | 142.2 to 146.7 | 144.4 | |
| 101 | Grok 4 Fast | xAI | Closed weights | 6 | 142.2 to 146.4 | 144.3 | |
| 102 | Gemma 4 31B IT | Google DeepMind | Open weights | 7 | 141.7 to 146.6 | 144.1 | |
| 103 | Qwen 3.6 35B-A3B | Alibaba | Open weights | 9 | 142.2 to 145.9 | 144.1 | |
| 104 | Qwen3.5-35B-A3B | Alibaba | Open weights | 4 | 141.5 to 146.6 | 144.1 | |
| 105 | Gemini 2.5 Pro (Mar 2025) | Google DeepMind | Closed weights | 10 | 142.0 to 145.9 | 144.0 | |
| 106 | Qwen 3.5 Flash (hosted 35B-A3B) | Alibaba | Other | 9 | 141.9 to 145.7 | 143.8 | |
| 107 | Qwen 3.6 Flash | Alibaba | Other | 10 | 141.8 to 145.6 | 143.7 | |
| 108 | GLM-4.7 | Z.ai (Zhipu AI) | Open weights | 11 | 141.4 to 145.3 | 143.3 | |
| 109 | Qwen3-235B-A22B-Thinking (Jul 2025) | Alibaba | Open weights | 10 | 141.3 to 145.1 | 143.2 | |
| 110 | Claude Opus 4 | Anthropic | Closed weights | 19 | 141.3 to 144.4 | 142.8 | |
| 111 | Claude Haiku 4.5 | Anthropic | Closed weights | 16 | 141.0 to 144.4 | 142.7 | |
| 112 | Qwen3-Max | Alibaba | Closed weights | 10 | 140.4 to 144.5 | 142.4 | |
| 113 | Gemini 2.5 Flash (Sep 2025) | Google DeepMind | Closed weights | 4 | 139.2 to 144.4 | 141.9 | |
| 114 | Claude Sonnet 4 | Anthropic | Closed weights | 21 | 140.3 to 143.3 | 141.8 | |
| 115 | Mistral Medium 3.5 | Mistral AI | Open weights | 6 | 138.3 to 144.7 | 141.6 | |
| 116 | Gemini 2.5 Flash (May 2025) | Google DeepMind | Closed weights | 7 | 139.4 to 143.6 | 141.6 | |
| 117 | Gemma 4 26B A4B | Google DeepMind | Open weights | 5 | 139.1 to 143.9 | 141.6 | |
| 118 | Kimi K2 (Jul 2025) | Moonshot | Open weights | 6 | 138.4 to 143.9 | 141.3 | |
| 119 | o1 | OpenAI | Closed weights | 17 | 139.5 to 142.8 | 141.2 | |
| 120 | Claude 3.7 Sonnet | Anthropic | Closed weights | 18 | 139.4 to 142.6 | 141.0 | |
| 121 | DeepSeek-V3.1 | DeepSeek | Open weights | 5 | 137.7 to 144.0 | 140.9 | |
| 122 | Gemini 2.5 Flash (Jun 2025) | Google DeepMind | Closed weights | 7 | 138.3 to 143.1 | 140.8 | |
| 123 | DeepSeek-R1 (May 2025) | DeepSeek | Open weights | 11 | 138.9 to 142.4 | 140.7 | |
| 124 | Gemini 2.5 Flash (Apr 2025) | Google DeepMind | Closed weights | 7 | 138.3 to 142.9 | 140.7 | |
| 125 | Gemini 2.5 Pro (May 2025) | Google DeepMind | Closed weights | 7 | 138.3 to 143.0 | 140.6 | |
| 125 | o3-mini | OpenAI | Closed weights | 19 | 138.9 to 142.3 | 140.6 | |
| 127 | gpt-oss-120b | OpenAI | Open weights | 13 | 138.3 to 142.0 | 140.2 | |
| 128 | Grok-3 mini | xAI | Closed weights | 9 | 138.0 to 141.7 | 139.9 | |
| 129 | Qwen3.5-9B | Alibaba | Open weights | 5 | 137.3 to 141.7 | 139.5 | |
| 130 | GPT-5 nano | OpenAI | Closed weights | 17 | 137.1 to 140.9 | 139.1 | |
| 131 | DeepSeek-R1 | DeepSeek | Open weights | 11 | 136.6 to 140.5 | 138.6 | |
| 132 | Qwen3-235B-A22B | Alibaba | Open weights | 8 | 136.0 to 140.7 | 138.4 | |
| 133 | Qwen3-30B-A3B-Thinking (Jul 2025) | Alibaba | Open weights | 4 | 135.8 to 140.8 | 138.3 | |
| 134 | Qwen3-235B-A22B-Instruct (Jul 2025) | Alibaba | Open weights | 7 | 135.7 to 140.2 | 138.0 | |
| 135 | Grok 3 | xAI | Closed weights | 12 | 135.6 to 139.4 | 137.6 | |
| 136 | Qwen3-32B | Alibaba | Open weights | 6 | 134.4 to 139.1 | 136.8 | |
| 137 | GPT-4.5 | OpenAI | Closed weights | 11 | 134.4 to 138.4 | 136.5 | |
| 138 | QwQ-32B | Alibaba | Open weights | 4 | 133.7 to 139.1 | 136.4 | |
| 139 | GPT-4.1 | OpenAI | Closed weights | 17 | 134.2 to 138.0 | 136.1 | |
| 140 | Qwen3-14B | Alibaba | Open weights | 5 | 133.3 to 138.7 | 136.1 | |
| 141 | Gemini 2.0 Pro | Google DeepMind | Closed weights | 5 | 132.6 to 138.0 | 135.4 | |
| 142 | o1-mini | OpenAI | Closed weights | 10 | 133.0 to 137.3 | 135.3 | |
| 143 | DeepSeek-V3 (Mar 2025) | DeepSeek | Open weights | 10 | 133.0 to 137.4 | 135.3 | |
| 144 | gpt-oss-20b | OpenAI | Open weights | 6 | 132.3 to 137.3 | 134.9 | |
| 145 | Qwen3-30B-A3B-Instruct (Jul 2025) | Alibaba | Open weights | 4 | 131.9 to 137.7 | 134.9 | |
| 146 | GPT-4.1 mini | OpenAI | Closed weights | 15 | 132.1 to 136.2 | 134.2 | |
| 147 | Qwen3-30B-A3B | Alibaba | Open weights | 6 | 131.3 to 136.7 | 134.2 | |
| 148 | Gemini 2.0 Flash (Feb 2025) | Google DeepMind,Google | Closed weights | 11 | 131.7 to 136.2 | 134.0 | |
| 149 | Gemini 2.0 Flash Thinking (Jan 2025) | Google DeepMind,Google | Closed weights | 6 | 130.7 to 136.4 | 133.6 | |
| 150 | o1-preview | OpenAI | Closed weights | 8 | 131.1 to 135.9 | 133.6 | |
| 151 | Claude 3.5 Sonnet (October 2024) | Anthropic | Closed weights | 17 | 131.2 to 135.5 | 133.4 | |
| 152 | Gemini 2.5 Flash-Lite (Jun 2025) | Google DeepMind | Closed weights | 4 | 129.3 to 136.7 | 133.0 | |
| 153 | Mistral Medium 3 | Mistral AI | Closed weights | 5 | 130.1 to 135.3 | 132.8 | |
| 154 | Llama 4 Maverick | Meta AI | Open weights | 13 | 129.8 to 134.3 | 132.2 | |
| 155 | Qwen3-8B | Alibaba | Open weights | 5 | 128.9 to 134.9 | 132.1 | |
| 156 | DeepSeek-V3 | DeepSeek | Open weights | 14 | 129.2 to 134.0 | 131.7 | |
| 157 | Qwen2.5-Max | Alibaba | Closed weights | 6 | 128.4 to 133.7 | 131.2 | |
| 158 | Gemini 1.5 Pro (Sept 2024) | Google DeepMind | Closed weights | 11 | 127.8 to 132.8 | 130.4 | |
| 159 | Claude 3.5 Sonnet | Anthropic | Closed weights | 12 | anchor, fixed | 130.0 | |
| 160 | Grok-2 (Dec 2024) | xAI | Closed weights | 7 | 127.0 to 132.3 | 129.8 | |
| 161 | Phi-4 | Microsoft Research | Open weights | 5 | 126.1 to 132.0 | 129.2 | |
| 162 | GPT-4o (Nov 2024) | OpenAI | Closed weights | 15 | 126.1 to 131.2 | 128.8 | |
| 163 | GPT-4o (May 2024) | OpenAI | Closed weights | 8 | 125.4 to 130.7 | 128.1 | |
| 164 | Llama 3.1-405B | Meta AI | Open weights | 15 | 125.0 to 130.5 | 127.9 | |
| 165 | Llama 4 Scout | Meta AI | Open weights | 9 | 124.9 to 130.3 | 127.8 | |
| 166 | Gemma 3 27B | Google DeepMind | Open weights | 8 | 124.4 to 130.4 | 127.6 | |
| 167 | Qwen2.5-72B | Alibaba | Open weights | 18 | 124.4 to 129.8 | 127.2 | |
| 168 | GPT-4o (Aug 2024) | OpenAI | Closed weights | 10 | 124.1 to 129.9 | 127.2 | |
| 169 | Gemini 1.5 Flash (Sep 2024) | Google DeepMind | Closed weights | 10 | 123.8 to 129.9 | 127.0 | |
| 170 | GPT-4.1 nano | OpenAI | Closed weights | 13 | 123.8 to 129.5 | 126.8 | |
| 171 | Mistral Large 2 (Jul 2024) | Mistral AI | Open weights | 6 | 123.1 to 129.4 | 126.4 | |
| 172 | Llama 3.3 70B | Meta AI | Open weights | 10 | 123.0 to 129.0 | 126.2 | |
| 173 | GPT-4 Turbo (Apr 2024) | OpenAI | Closed weights | 8 | 122.5 to 128.8 | 125.8 | |
| 174 | Claude 3 Opus | Anthropic | Closed weights | 13 | 122.5 to 128.3 | 125.6 | |
| 175 | Gemini 1.5 Pro (May 2024) | Google DeepMind | Closed weights | 6 | 122.1 to 128.3 | 125.4 | |
| 176 | Claude 3.5 Haiku | Anthropic | Closed weights | 11 | 121.5 to 127.7 | 124.8 | |
| 176 | GPT-4 Turbo (Nov 2023) | OpenAI | Closed weights | 6 | 121.5 to 127.7 | 124.8 | |
| 178 | Llama 3.1-70B | Meta AI | Open weights | 9 | 121.4 to 127.7 | 124.7 | |
| 179 | GPT-4o mini | OpenAI | Closed weights | 19 | 120.5 to 126.8 | 123.8 | |
| 180 | Llama 3.2 90B | Meta AI | Open weights | 6 | 119.2 to 126.3 | 122.9 | |
| 181 | Qwen2-72B | Alibaba | Open weights | 6 | 118.6 to 125.6 | 122.3 | |
| 182 | GPT-4 (Mar 2023) | OpenAI | Closed weights | 7 | 117.6 to 125.2 | 121.6 | |
| 183 | Llama 3-70B | Meta AI | Open weights | 7 | 117.5 to 124.7 | 121.3 | |
| 184 | DeepSeek-V2 (MoE-236B, May 2024) | DeepSeek | Open weights | 7 | 114.0 to 127.6 | 120.8 | |
| 185 | Gemini 1.5 Flash (May 2024) | Google DeepMind | Closed weights | 6 | 116.7 to 124.1 | 120.6 | |
| 186 | Mistral Large | Mistral AI | Closed weights | 5 | 115.9 to 123.6 | 120.0 | |
| 187 | GPT-4 (Jun 2023) | OpenAI | Closed weights | 11 | 116.2 to 123.3 | 119.9 | |
| 188 | Claude 3 Sonnet | Anthropic | Closed weights | 7 | 115.5 to 123.2 | 119.7 | |
| 189 | Qwen2.5-Coder-32B | Alibaba | Open weights | 5 | 109.5 to 127.8 | 118.7 | |
| 190 | Mixtral 8x22B | Mistral AI | Open weights | 6 | 114.5 to 122.5 | 118.7 | |
| 191 | Gemma 2 27B | Google DeepMind | Open weights | 6 | 114.2 to 122.4 | 118.6 | |
| 192 | Claude 3 Haiku | Anthropic | Closed weights | 9 | 113.5 to 121.5 | 117.8 | |
| 193 | Claude 2 | Anthropic | Closed weights | 6 | 112.8 to 121.2 | 117.2 | |
| 194 | phi-3-small 7.4B | Microsoft | Open weights | 8 | 110.5 to 121.9 | 116.3 | |
| 195 | phi-3-medium 14B | Microsoft | Open weights | 10 | 109.2 to 118.5 | 114.1 | |
| 196 | Claude Instant | Anthropic | Closed weights | 5 | 107.0 to 120.3 | 114.0 | |
| 197 | PaLM 2-L | not stated | Other | 4 | 98.4 to 130.4 | 114.0 | |
| 198 | Mistral NeMo | Mistral AI | Open weights | 6 | 108.6 to 118.4 | 113.9 | |
| 199 | Qwen2.5-Coder-14B | not stated | Other | 5 | 105.2 to 122.3 | 113.7 | |
| 200 | Mixtral 8x7B | Mistral AI | Open weights | 12 | 108.1 to 117.5 | 113.2 | |
| 201 | GPT-3.5 Turbo (Jan 2024) | OpenAI | Closed weights | 9 | 107.8 to 117.3 | 112.9 | |
| 202 | GPT-3.5 Turbo (Nov 2023) | OpenAI | Closed weights | 8 | 107.3 to 117.2 | 112.7 | |
| 203 | Gemma 2 9B | Google DeepMind | Open weights | 6 | 106.8 to 117.3 | 112.4 | |
| 204 | Llama 3.1-8B | Meta AI | Open weights | 10 | 105.1 to 115.9 | 110.9 | |
| 205 | phi-3-mini 3.8B | Microsoft | Open weights | 8 | 104.0 to 116.1 | 110.5 | |
| 206 | Stable Beluga 2 | Stability AI | Open weights | 7 | 102.0 to 116.1 | 109.4 | |
| 207 | Llama 3-8B | Meta AI | Open weights | 10 | 103.4 to 114.0 | 109.1 | |
| 208 | Qwen2.5-Coder (7B) | Alibaba | Open weights | 5 | 97.7 to 116.1 | 107.4 | |
| 209 | Falcon-180B | Technology Innovation Institute | Open weights | 8 | 98.2 to 115.5 | 107.0 | |
| 210 | GPT-3.5 Turbo (Jun 2023) | OpenAI | Closed weights | 4 | 97.6 to 113.4 | 106.0 | |
| 211 | Llama 2-70B | Meta AI | Open weights | 14 | 99.7 to 111.0 | 105.7 | |
| 212 | Gemma 7B | Google DeepMind | Open weights | 10 | 95.4 to 109.7 | 103.2 | |
| 213 | Mistral 7B v0.1 | Mistral AI | Open weights | 10 | 94.9 to 109.1 | 102.7 | |
| 214 | Yi-9B | not stated | Other | 4 | 91.3 to 111.8 | 102.2 | |
| 215 | Qwen-14B | Alibaba | Open weights | 6 | 93.2 to 109.3 | 101.8 | |
| 216 | LLaMA-65B | Meta AI | Open weights | 10 | 93.2 to 108.9 | 101.5 | |
| 217 | DeepSeek-Coder-V2-Lite-Base | not stated | Other | 4 | 88.8 to 108.4 | 99.4 | |
| 218 | Falcon 2 11B | Technology Innovation Institute | Open weights | 4 | 88.4 to 109.0 | 99.2 | |
| 219 | Nemotron-4 15B | Nvidia | Closed weights | 7 | 90.0 to 107.0 | 99.1 | |
| 220 | internlm-20b | not stated | Other | 6 | 89.1 to 107.5 | 98.9 | |
| 221 | PaLM 2-M | not stated | Other | 5 | 84.6 to 109.7 | 97.9 | |
| 222 | Qwen2.5-Coder-3B | not stated | Other | 4 | 82.9 to 109.1 | 96.8 | |
| 223 | LLaMA-33B | Meta AI | Open weights | 10 | 87.4 to 104.5 | 96.5 | |
| 224 | Llama 2-13B | Meta AI | Open weights | 11 | 87.4 to 104.2 | 96.5 | |
| 225 | Phi-2 | Microsoft | Open weights | 8 | 86.9 to 104.3 | 96.2 | |
| 226 | StarCoder 2 15B | Hugging Face,ServiceNow,Nvidia,BigCode | Open weights | 4 | 84.0 to 106.0 | 96.1 | |
| 227 | Yi 6B | 01.AI | Open weights | 6 | 84.1 to 103.2 | 94.3 | |
| 228 | Llama 2-34B | Meta AI | Closed weights | 8 | 83.8 to 102.6 | 93.9 | |
| 229 | PaLM 2-S | not stated | Other | 5 | 80.7 to 105.7 | 93.8 | |
| 230 | Baichuan2-13B | Baichuan | Open weights | 7 | 80.8 to 100.9 | 91.8 | |
| 231 | Falcon-40B | Technology Innovation Institute | Open weights | 10 | 80.0 to 100.0 | 90.7 | |
| 232 | Qwen-7B | Alibaba | Open weights | 6 | 77.9 to 98.7 | 89.1 | |
| 233 | Qwen2.5-Coder (1.5B) | Alibaba | Open weights | 5 | 76.0 to 100.1 | 89.1 | |
| 234 | LLaMA-13B | Meta AI | Open weights | 10 | 73.4 to 95.4 | 85.2 | |
| 235 | MPT-30B | MosaicML | Open weights | 8 | 73.3 to 95.3 | 85.2 | |
| 236 | internlm-7b | not stated | Other | 7 | 72.5 to 94.8 | 84.6 | |
| 237 | Baichuan 2-7B | Baichuan | Open weights | 6 | 70.5 to 94.5 | 83.6 | |
| 238 | INTELLECT-1 | Prime Intellect,Hugging Face,Arcee AI | Other | 6 | 70.8 to 94.0 | 83.3 | |
| 239 | Llama 2-7B | Meta AI | Open weights | 10 | 70.6 to 93.8 | 83.2 | |
| 240 | LLaMA-7B | Meta AI | Open weights | 10 | 63.4 to 88.9 | 77.3 | |
| 241 | DeepSeek Coder 33B | DeepSeek,Peking University | Open weights | 4 | 60.4 to 89.5 | 76.3 | |
| 242 | vicuna-13b-v1.1 | not stated | Other | 7 | 60.8 to 88.6 | 76.1 | |
| 243 | Gemma 2B | Google DeepMind | Open weights | 8 | 61.0 to 87.6 | 75.7 | |
| 244 | CodeQwen1.5-7B | not stated | Other | 4 | 57.5 to 88.6 | 74.8 | |
| 245 | chatglm2-6b | not stated | Other | 7 | 58.4 to 86.8 | 73.7 | |
| 246 | MPT-7B | MosaicML | Open weights | 10 | 58.2 to 86.2 | 73.4 | |
| 247 | Falcon-7B | Technology Innovation Institute | Open weights | 10 | 58.2 to 85.4 | 73.1 | |
| 248 | StarCoder 2 7B | Hugging Face,ServiceNow,Nvidia,BigCode | Open weights | 4 | 54.9 to 86.6 | 72.4 | |
| 249 | XGen-7B | Salesforce | Open weights | 6 | 54.3 to 84.2 | 70.3 | |
| 250 | Phi-1.5 | Microsoft | Open weights | 5 | 50.5 to 82.9 | 68.4 | |
| 251 | Qwen2.5-Coder-0.5B | not stated | Other | 5 | 48.4 to 83.9 | 68.1 | |
| 252 | DeepSeek Coder 6.7B | DeepSeek,Peking University | Open weights | 4 | 49.1 to 82.2 | 67.5 | |
| 253 | StarCoder 2 3B | Hugging Face,ServiceNow,Nvidia,BigCode | Open weights | 4 | 47.9 to 81.7 | 66.6 | |
| 254 | open_llama_7b | not stated | Other | 6 | 46.0 to 79.5 | 64.4 | |
| 255 | Qwen-1_8B | not stated | Other | 6 | 40.2 to 75.7 | 59.6 | |
| 256 | RedPajama-INCITE-7B-Base | not stated | Other | 6 | 37.9 to 74.4 | 57.7 | |
| 257 | Dolly 2.0-12b | Databricks | Open weights | 6 | 34.5 to 72.6 | 55.5 | |
| 258 | Cerebras-GPT-13B | Cerebras Systems | Open weights | 6 | 26.3 to 67.1 | 48.9 |
258 of 258 models, highest index first. A model links to its catalog page where the pairing names one, which is true of 116; the rest are models the catalog does not list, and the record Epoch keeps of them stands on its own. Results is how many benchmark results stand behind the figure, and Epoch places nothing on fewer than 4. The two rows with no interval are the anchors the scale is defined by, whose figures are fixed rather than estimated.
Source: the Epoch AI Capabilities Index (epoch.ai/eci), as of 20 Nov 2026. 258 models placed over 2917 benchmark results.
Epoch AI Benchmarking Hub data, by Epoch AI, under CC BY 4.0.
Epoch states that absolute values are meaningless by themselves and that comparisons between models are what the index is for, that values are not linearly related to benchmark accuracy, and that a highly specialised model may score low while being capable in its own domain.
The index is Epoch's fit. Vitna ran none of these evaluations, computed none of these figures and adjusted none of them.
