Epoch Capabilities Index leaderboard
Epoch Capabilities Index leaderboard: one intelligence score per LLM that combines 50+ benchmarks, plotted against API price to compare overall capability and cost. Current leader: Claude Opus 5.5 at 168.8 (ECI).
265 models
Top score
Claude Opus 5.5 leads Epoch Capabilities Index with 168.8 at $20.0 per 1M output tokens. GPT-6 Astra follows at 168.8.
Best value
GPT-6.1 Sol scores 168.4, 0.4 points off the leader, for $10.0 per 1M output tokens, about 2.0x cheaper.
Cheapest strong model
Qwen3.7 Flash is the cheapest model scoring 140+ (144.1), at $0.13 per 1M output tokens.
Best open-weight
Kimi K3 is the best open-weight model at 157.9, 10.9 points behind the leader, at $11.9 per 1M output tokens.
Cost vs quality
Epoch Capabilities Index: score against cost
Each dot is a model. Top-left is best: a high score for a low cost. The dashed line is the value frontier.
Score (ECI)
$0.1$1$10$100$1,000
Output price per 1M tokens, USD (log scale)
Closed modelOpen-weight modelValue frontier63 of 265 models have a cost and appear here.
Epoch Capabilities Index leaderboard
| Model | ECI | Output $ / 1M | Released |
|---|---|---|---|
1 Claude Opus 5.5 | 168.8 | $20.0 | Sep 2026 |
2 GPT-6 Astra | 168.8 | $50.0 | Sep 2026 |
| 3GPT-6.1 Sol | 168.4 | $10.0 | Sep 2026 |
| 4Claude Fable 5.1 | 166.5 | $50.0 | Sep 2026 |
5 Gemini 4 Argon | 166.4 | $10.0 | Sep 2026 |
| 6Claude Sonnet 5.5 | 166.2 | $10.0 | Sep 2026 |
| 7Claude Opus 5 | 163.9 | $25.0 | Jul 2026 |
| 8GPT-6 Sol | 163.7 | $10.0 | Sep 2026 |
| 9Claude Fable 5 | 163.4 | $50.0 | Jun 2026 |
| 10GPT-5.5 Pro | 162.9 | $180 | Apr 2026 |
| 11GPT-5.6 Sol | 162.8 | $20.0 | Jul 2026 |
| 12Muse Wonder | 161.8 | — | Oct 2026 |
| 13GPT-5.6 Terra | 160.1 | $12.0 | Jul 2026 |
| 14GPT-5.5 | 159.7 | $30.0 | Apr 2026 |
| 15GPT-5.4 Pro | 159.6 | $180 | Mar 2026 |
| 16Claude Haiku 5.5 | 159.3 | $0.50 | Oct 2026 |
| 17Claude Opus 4.8 | 159 | $25.0 | May 2026 |
| 18Gemini 3.7 Flash | 158 | $3.75 | Aug 2026 |
19 Kimi K3Open | 157.9 | $11.9 | Jul 2026 |
| 20Gemini 3.8 Flash | 157.4 | $4.13 | Sep 2026 |
21 Qwen 3.8 Max | 157.4 | — | Jul 2026 |
| 22GPT-5.4 | 157.3 | $15.0 | Mar 2026 |
| 23Muse Spark 1.3 | 157.2 | — | Sep 2026 |
| 24GPT-5.3 Codex | 157.1 | $14.0 | Feb 2026 |
| 25Claude Sonnet 5 | 156.9 | $10.0 | Jun 2026 |
| 26Claude Opus 4.7 | 156.8 | $25.0 | Apr 2026 |
| 27GPT-6 Luna | 156.8 | $0.50 | Sep 2026 |
28 Grok 4.6 | 156.6 | $6.00 | Aug 2026 |
| 29GPT-5.6 Luna | 156.5 | $1.20 | Jul 2026 |
| 30GPT-5.2 Pro | 156.3 | — | Dec 2025 |
31 GLM-5.3Open | 156 | $1.40 | Aug 2026 |
32 DeepSeek V4.1 FlashOpen | 155.9 | $1.20 | Sep 2026 |
| 33Claude Opus 4.6 | 155.7 | $25.0 | Feb 2026 |
| 34DeepSeek V4 Pro 0813 | 155.6 | $1.98 | Aug 2026 |
| 35Gemini 3.1 Pro | 155.3 | $12.0 | Feb 2026 |
| 36Muse Spark OpenOpen | 155.1 | — | Oct 2026 |
| 37Qwen3.8 Max (0902) | 155.1 | $6.00 | Sep 2026 |
| 38Muse Spark 1.2 | 155.1 | — | Aug 2026 |
| 39Gemini 3.5 Flash | 154.9 | $9.00 | May 2026 |
| 40DeepSeek V4 Flash 0731Open | 154.8 | $1.32 | Jul 2026 |
| 41Gemini 3.6 Flash | 154.5 | $3.75 | Jul 2026 |
| 42Grok 4.7 | 154.4 | $6.00 | Sep 2026 |
| 43Grok 4.5 | 154.1 | $6.00 | Jul 2026 |
| 44Muse Spark 1.1 | 154.1 | — | Jul 2026 |
| 45Qwen3.7-Max | 154 | $4.42 | May 2026 |
| 46GPT-5.2 | 153.7 | — | Dec 2025 |
| 47GLM-5.3-FlashOpen | 153.1 | $0.20 | Aug 2026 |
| 48Gemini 3 Pro | 153.1 | — | Nov 2025 |
| 49Claude Sonnet 4.6 | 152.8 | $15.0 | Feb 2026 |
| 50Grok 4.20 | 152.1 | $2.50 | Feb 2026 |
| 51Muse Spark | 152 | — | Apr 2026 |
| 52GLM-5.2Open | 151.9 | $7.00 | Jun 2026 |
| 53Gemini 3 Flash | 151.9 | — | Dec 2025 |
| 54Qwen 3.8 27BOpen | 151.7 | — | Aug 2026 |
| 55Kimi K2.6Open | 150.9 | $3.41 | Apr 2026 |
| 56GPT-5 Pro | 150.6 | — | Oct 2025 |
| 57Inkling-SmallOpen | 150.3 | — | Jul 2026 |
| 58Kimi K2.7 CodeOpen | 150.1 | $3.35 | Jun 2026 |
| 59Claude Opus 4.5 | 150.1 | — | Nov 2025 |
60 Mistral Large 4 | 150 | $4.18 | Oct 2026 |
| 61GPT-5 | 150 | — | Aug 2025 |
| 62GLM-5.1Open | 149.8 | $4.40 | Apr 2026 |
| 63GPT-5.1 | 149.8 | — | Nov 2025 |
| 64Qwen 3.6 Max (Preview) | 149.4 | — | Apr 2026 |
| 65DeepSeek-V4-ProOpen | 149.1 | $1.91 | Apr 2026 |
| 66Grok 4.3 Beta | 148.8 | — | Apr 2026 |
| 67GPT-5.4 Mini | 148.7 | $4.50 | Mar 2026 |
| 68InklingOpen | 148.4 | — | Jul 2026 |
| 69Kimi K2.5Open | 148.1 | $2.50 | Feb 2026 |
| 70o3-pro | 147.3 | — | Jun 2025 |
| 71Qwen 3.6 Plus | 147.3 | — | Apr 2026 |
| 72Qwen3.7-Plus | 147 | $1.28 | Jun 2026 |
73 MiniMax-M3Open | 146.9 | — | Jun 2026 |
| 74Qwen3.5 397B-A17BOpen | 146.8 | $3.60 | Feb 2026 |
| 75Claude Sonnet 4.5 | 146.7 | — | Sep 2025 |
| 76o3 | 146.7 | — | Dec 2024 |
| 77MiniMax-M2.5Open | 146.6 | — | Feb 2026 |
| 78Grok 4 | 146.2 | — | Sep 2025 |
| 79Qwen3.6 27BOpen | 146.2 | $2.00 | Apr 2026 |
| 80Qwen 3.5 Plus (hosted 397B-A17B) | 146.2 | — | Feb 2026 |
| 81DeepSeek-V4-FlashOpen | 146.1 | $0.18 | Apr 2026 |
| 82DeepSeek-V3.2Open | 146.1 | — | Dec 2025 |
| 83GPT-5.4 Nano | 146 | $1.25 | Mar 2026 |
84 Nemotron 3 UltraOpen | 146 | — | Jun 2026 |
| 85Kimi K2 ThinkingOpen | 145.8 | — | Nov 2025 |
| 86GLM-5Open | 145.8 | $3.20 | Feb 2026 |
| 87MiniMax-M2.7Open | 145.8 | — | Mar 2026 |
| 88o4-mini | 145.5 | — | Apr 2025 |
| 89GPT-5 mini | 145.3 | — | Aug 2025 |
| 90Gemini 2.5 Pro (Jun 2025) | 145.1 | — | Jun 2025 |
| 91Gemini 3.5 Flash-Lite | 144.9 | $2.50 | Jul 2026 |
| 92Gemini 3.1 Flash-Lite | 144.3 | $1.50 | Mar 2026 |
| 93Claude Opus 4.1 | 144.1 | — | Aug 2025 |
| 94Qwen3.7 Flash | 144.1 | $0.13 | Jul 2026 |
| 95Gemini 2.5 Pro (Mar 2025) | 144 | — | Mar 2025 |
| 96Qwen3-235B-A22B-Thinking (Jul 2025)Open | 143.8 | — | Jul 2025 |
| 97Qwen 3.6 35B-A3B | 143.7 | — | Apr 2026 |
| 98GLM-4.7Open | 143.6 | — | Dec 2025 |
| 99Qwen 3.5 Flash (hosted 35B-A3B) | 143.5 | — | Feb 2026 |
| 100Gemma 4 31B ITOpen | 143 | $0.40 | Apr 2026 |
| 101Gemini 2.5 Flash (Sep 2025) | 142.9 | — | Sep 2025 |
| 102Qwen 3.6 Flash | 142.9 | — | Apr 2026 |
| 103Claude Opus 4 | 142.5 | — | May 2025 |
| 104Qwen3.5-35B-A3BOpen | 142.5 | $1.00 | Feb 2026 |
| 105GPT-5.5 Instant | 142.4 | — | May 2026 |
| 106Qwen3-Max | 142.3 | — | Jan 2026 |
| 107Claude Haiku 4.5 | 142.2 | — | Oct 2025 |
| 108Gemini 2.5 Pro (May 2025) | 142.2 | — | May 2025 |
| 109Gemma 4 26B A4BOpen | 141.9 | $0.30 | Apr 2026 |
| 110o1 | 141.7 | — | Dec 2024 |
| 111Claude Sonnet 4 | 141.6 | — | May 2025 |
| 112Gemini 2.5 Flash (May 2025) | 141.4 | — | May 2025 |
| 113Mistral Medium 3.5Open | 141.4 | $7.50 | Apr 2026 |
| 114DeepSeek-R1 (May 2025)Open | 141.1 | — | May 2025 |
| 115Claude 3.7 Sonnet | 141 | — | Feb 2025 |
| 116Gemini 2.5 Flash (Jun 2025) | 140.6 | — | Jun 2025 |
| 117o3-mini | 140.2 | — | Jan 2025 |
| 118Grok-3 mini | 140.2 | — | Feb 2025 |
| 119Kimi K2 (Jul 2025) | 140 | — | Jul 2025 |
| 120Gemini 2.5 Flash (Apr 2025) | 139.8 | — | Apr 2025 |
| 121gpt-oss-120bOpen | 139.8 | — | Aug 2025 |
| 122DeepSeek-V3.1Open | 139.8 | — | Aug 2025 |
| 123Qwen3-30B-A3B-Thinking (Jul 2025)Open | 139.6 | — | Jul 2025 |
| 124Qwen3.5-9BOpen | 139.3 | $0.25 | Feb 2026 |
| 125GPT-5 nano | 139.2 | — | Aug 2025 |
| 126Qwen3-235B-A22BOpen | 139.2 | — | Apr 2025 |
| 127DeepSeek-R1Open | 138.8 | — | Jan 2025 |
| 128Qwen3-32BOpen | 138.4 | — | Apr 2025 |
| 129Grok 3 | 138.2 | — | Feb 2025 |
| 130Qwen3-14BOpen | 138.2 | — | Apr 2025 |
| 131gpt-oss-20bOpen | 137.7 | — | Aug 2025 |
132 QwQ-32BOpen | 137.5 | — | Mar 2025 |
| 133GPT-4.5 | 136.7 | — | Feb 2025 |
| 134GPT-4.1 | 136.7 | — | Apr 2025 |
| 135Qwen3-30B-A3BOpen | 136.1 | — | Apr 2025 |
| 136Qwen3-8BOpen | 136.1 | — | Apr 2025 |
| 137DeepSeek-V3 (Mar 2025)Open | 135.8 | — | Mar 2025 |
| 138o1-mini | 135.7 | — | Sep 2024 |
| 139Gemini 2.0 Flash Thinking (Jan 2025) | 135.2 | — | Jan 2025 |
| 140Gemini 2.0 Pro | 134.9 | — | Dec 2024 |
| 141GPT-4.1 mini | 134.9 | — | Apr 2025 |
| 142Gemini 2.0 Flash (Feb 2025) | 134.6 | — | Feb 2025 |
| 143Gemini 2.0 Flash (Dec 2024) | 134.6 | — | Dec 2024 |
| 144Mistral Medium 3 | 134 | — | May 2025 |
| 145Gemini 2.5 Flash-Lite (Jun 2025) | 133.8 | — | Jun 2025 |
| 146Claude 3.5 Sonnet (October 2024) | 133.4 | — | Oct 2024 |
| 147Magistral Small 1.0Open | 133.1 | — | Jun 2025 |
| 148Qwen2.5-Max | 132.5 | — | Jan 2025 |
| 149DeepSeek-V3Open | 132.3 | — | Dec 2024 |
150 Llama 4 MaverickOpen | 132.2 | — | Apr 2025 |
| 151Gemini 1.5 Pro (Sept 2024) | 131.7 | — | Sep 2024 |
152 Phi-4Open | 130.5 | — | Apr 2025 |
| 153Grok-2 (Dec 2024) | 130.5 | — | Dec 2024 |
| 154Claude 3.5 Sonnet | 130 | — | Jun 2024 |
| 155Gemma 3 27BOpen | 130 | — | Mar 2025 |
| 156Llama 4 ScoutOpen | 129.6 | — | Apr 2025 |
| 157GPT-4.1 nano | 129.5 | — | Apr 2025 |
| 158Gemini 1.5 Flash (Sep 2024) | 129.3 | — | Sep 2024 |
| 159GPT-4o (May 2024) | 129 | — | May 2024 |
| 160Qwen2.5-72BOpen | 129 | — | Sep 2024 |
| 161GPT-4o (Aug 2024) | 128.8 | — | Aug 2024 |
| 162Llama 3.1-405BOpen | 128.8 | — | Jul 2024 |
| 163GPT-4o (Nov 2024) | 128.7 | — | Nov 2024 |
| 164Mistral Large 2 (Nov 2024) | 128.6 | — | Nov 2024 |
| 165Mistral Large 2 (Jul 2024) | 127.6 | — | Jul 2024 |
| 166Mistral Small 3.1Open | 127.5 | — | Mar 2025 |
| 167Llama 3.3 70BOpen | 127.3 | — | Dec 2024 |
| 168GPT-4 Turbo (Apr 2024) | 127.3 | — | Apr 2024 |
| 169Claude 3.5 Haiku | 127.1 | — | Oct 2024 |
| 170Gemini 1.5 Pro (May 2024) | 126.9 | — | May 2024 |
| 171Claude 3 Opus | 126.9 | — | Mar 2024 |
| 172GPT-4o mini | 126.5 | — | Jul 2024 |
| 173GPT-4 Turbo (Nov 2023) | 126.4 | — | Nov 2023 |
| 174Llama 3.1-70BOpen | 125.9 | — | Jul 2024 |
| 175GPT-4 (Mar 2023) | 125.7 | — | Mar 2023 |
| 176Llama 3.2 90BOpen | 125.5 | — | Sep 2024 |
| 177Qwen2-72BOpen | 125.3 | — | Jun 2024 |
| 178DeepSeek-V2 (MoE-236B, May 2024) | 124.8 | — | May 2024 |
| 179Gemma 3 12BOpen | 123.5 | — | Mar 2025 |
| 180GPT-4 (Jun 2023) | 123.1 | — | Jun 2023 |
| 181Llama 3-70BOpen | 123 | — | Apr 2024 |
| 182Gemini 1.5 Flash (May 2024) | 122.7 | — | May 2024 |
| 183Mistral Large | 122.1 | — | Feb 2024 |
| 184Gemma 2 27BOpen | 122.1 | — | Jun 2024 |
| 185Mixtral 8x22BOpen | 122 | — | Apr 2024 |
| 186phi-3-small 7.4BOpen | 121.8 | — | Apr 2024 |
| 187phi-3-medium 14BOpen | 121.2 | — | Apr 2024 |
| 188Claude 3 Sonnet | 120.8 | — | Mar 2024 |
| 189Claude Instant | 120.2 | — | Aug 2023 |
| 190Claude 2 | 120.2 | — | Jul 2023 |
| 191Gemma 2 9BOpen | 119.8 | — | Jun 2024 |
| 192Qwen2.5-Coder-32BOpen | 119.5 | — | Nov 2024 |
| 193Claude 2.1 | 119.3 | — | Nov 2023 |
194 Command R+Open | 119.2 | — | Apr 2024 |
| 195Claude 3 Haiku | 118.8 | — | Mar 2024 |
| 196Mistral NeMoOpen | 118.7 | — | Jul 2024 |
| 197Qwen2.5-7BOpen | 118.5 | — | Sep 2024 |
| 198GPT-3.5 Turbo (Nov 2023) | 118.5 | — | Nov 2023 |
| 199Mixtral 8x7BOpen | 118.4 | — | Dec 2023 |
| 200Ministral 3B | 118 | — | Oct 2024 |
| 201Yi-34BOpen | 117.3 | — | Nov 2023 |
| 202phi-3-mini 3.8BOpen | 117.3 | — | Apr 2024 |
| 203Gemini 1.0 Pro | 117.1 | — | Dec 2023 |
| 204Stable Beluga 2Open | 117 | — | Jul 2023 |
| 205Llama 3.1-8BOpen | 116.5 | — | Jul 2024 |
| 206Llama 3-8BOpen | 116.4 | — | Apr 2024 |
| 207Qwen2.5-Coder-14B | 116.3 | — | Sep 2024 |
| 208Gemma 3 4BOpen | 115.9 | — | Mar 2025 |
| 209GPT-3.5 Turbo (Jan 2024) | 115.7 | — | Jan 2024 |
| 210PaLM 2-L | 114.9 | — | May 2023 |
| 211Llama 2-70BOpen | 113.7 | — | Jul 2023 |
| 212GPT-3.5 Turbo (Jun 2023) | 113.2 | — | Jun 2023 |
| 213Qwen2.5-Coder (7B)Open | 113 | — | Sep 2024 |
| 214Qwen-14BOpen | 112.9 | — | Sep 2023 |
| 215Mistral 7B v0.1 | 112 | — | Sep 2023 |
| 216Falcon-180BOpen | 112 | — | Sep 2023 |
| 217internlm-20b | 111.9 | — | Sep 2023 |
| 218Gemma 7BOpen | 111.8 | — | Feb 2024 |
| 219LLaMA-65BOpen | 110 | — | Feb 2023 |
| 220Falcon 2 11BOpen | 109.4 | — | May 2024 |
| 221Mistral 7B v0.3 | 108.9 | — | May 2024 |
| 222DeepSeek-Coder-V2-Lite-Base | 108.7 | — | Jun 2024 |
| 223PaLM 2-M | 108.1 | — | May 2023 |
| 224Phi-2Open | 107.7 | — | Dec 2023 |
| 225Qwen2.5-Coder-3B | 107.5 | — | Sep 2024 |
| 226Nemotron-4 15B | 107.5 | — | Feb 2024 |
| 227Yi-9B | 107.4 | — | Mar 2024 |
| 228LLaMA-33BOpen | 107.2 | — | Feb 2023 |
| 229Qwen-7BOpen | 106.6 | — | Sep 2023 |
| 230PaLM 2-S | 105.9 | — | May 2023 |
| 231Llama 2-13BOpen | 105.6 | — | Jul 2023 |
| 232Llama 2-34B | 104.9 | — | Jul 2023 |
| 233StarCoder 2 15BOpen | 104.7 | — | Feb 2024 |
| 234Yi 6BOpen | 104.5 | — | Nov 2023 |
| 235Falcon-40BOpen | 104.2 | — | Mar 2023 |
| 236Baichuan2-13BOpen | 102.9 | — | Sep 2023 |
| 237Qwen2.5-Coder (1.5B)Open | 102.7 | — | Sep 2024 |
| 238internlm-7b | 102.6 | — | Jul 2023 |
| 239INTELLECT-1 | 100.5 | — | Nov 2024 |
| 240MPT-30BOpen | 100.3 | — | Jun 2023 |
| 241LLaMA-13BOpen | 100.3 | — | Feb 2023 |
| 242Llama 2-7BOpen | 98.6 | — | Jul 2023 |
| 243chatglm2-6b | 98.6 | — | Jun 2023 |
| 244LLaMA-7BOpen | 96.6 | — | Feb 2023 |
| 245Baichuan 2-7BOpen | 95.9 | — | Sep 2023 |
| 246DeepSeek Coder 33BOpen | 95.9 | — | Nov 2023 |
| 247Falcon-7BOpen | 94.7 | — | Apr 2023 |
| 248CodeQwen1.5-7B | 94.5 | — | Apr 2024 |
| 249vicuna-13b-v1.1 | 94.3 | — | Apr 2023 |
| 250MPT-7BOpen | 94.2 | — | May 2023 |
| 251Gemma 2BOpen | 93.8 | — | Feb 2024 |
| 252StarCoder 2 7BOpen | 93.1 | — | Feb 2024 |
| 253XGen-7BOpen | 93 | — | Sep 2023 |
| 254Qwen-1_8B | 92.5 | — | Nov 2023 |
| 255open_llama_7b | 91.2 | — | Jun 2023 |
| 256Phi-1.5Open | 91.1 | — | Sep 2023 |
| 257Baichuan1-7BOpen | 90 | — | Jun 2023 |
| 258RedPajama-INCITE-7B-BaseOpen | 90 | — | Jun 2023 |
| 259Dolly 2.0-12bOpen | 89.2 | — | Apr 2023 |
| 260DeepSeek Coder 6.7BOpen | 89.1 | — | Nov 2023 |
| 261StarCoder 2 3BOpen | 88.3 | — | Feb 2024 |
| 262Qwen2.5-Coder-0.5B | 87.6 | — | Sep 2024 |
| 263Cerebras-GPT-13BOpen | 82.7 | — | Apr 2023 |
| 264DeepSeek Coder 1.3BOpen | 62.6 | — | Nov 2023 |
| 265stablelm-tuned-alpha-7b | 54.8 | — | Apr 2023 |
Scores from Epoch AI, updated —. Cost is joined from our daily API pricing where a model matches; models we do not price still appear with their score. Always verify on the benchmark's own site before relying on a number.
What the Epoch Capabilities Index measures
The Epoch Capabilities Index (ECI) is a single score for how capable a model is overall. Epoch AI fits it from more than 50 benchmark results using a statistical model that accounts for how hard each benchmark is, so it keeps separating models after any one test has saturated. The scale is anchored so that Claude 3.5 Sonnet scores 130 and GPT-5 scores 150. A model needs results on at least four benchmarks to get a score.
Why we use it as our intelligence measure
A single benchmark such as GPQA Diamond tops out near 100% and bunches the best models together. ECI blends many benchmarks, so it is harder to game and still ranks frontier models. Each score comes with a confidence interval; when two models' intervals overlap, treat them as statistically tied rather than reading meaning into a point or two.
ECI is published by Epoch AI under CC BY 4.0 and refreshed daily. We show it next to each model's price; GPQA Diamond and the other benchmarks remain available on their own pages.
Frequently asked questions
What is the Epoch Capabilities Index?
A composite intelligence score from Epoch AI that combines more than 50 benchmarks into one number per model. Claude 3.5 Sonnet is 130 and GPT-5 is 150.
Why not just use GPQA Diamond?
GPQA Diamond is saturated: top models score 97 to 99, so it can no longer tell them apart. ECI keeps separating frontier models.
What does the confidence interval mean?
It is the range Epoch AI considers plausible for the true score. If two models' ranges overlap, they are statistically tied.