01 Data & objectives
Curated data and learning objectives shape training. Data rights, coverage and quality require review.
Model companies and open communities build LLMs and other models. Harnesses connect them to context and tools; agent systems organize their capabilities into useful work.
A model is developed, evaluated and served. Tools and context can extend what it does; testing determines where it is useful.
Original conceptual illustration, not a site plan or engineering specification. Arrows show selected relationships, not measured flows or capacity. Numbered components are explained below; select one in the diagram to jump to it.
Curated data and learning objectives shape training. Data rights, coverage and quality require review.
Compute updates model parameters; further training can adapt behavior and task performance.
Capability, reliability and safety evaluations test a particular model and deployment configuration.
Serving systems execute the model for new inputs, trading off latency, throughput and cost.
Retrieval supplies relevant information; tools let a system interact with software under defined permissions.
An API or product delivers capabilities to a workflow. Production feedback informs further evaluation.
Technical context: NVIDIA: AI is a 5-layer cake ↗ · Anthropic: Building effective agents ↗
Epoch’s Capabilities Index brings multiple benchmarks onto a common scale. Follow model releases across organizations and open- or closed-weight availability.
ECI points · model release date · all scores from one dataset snapshot
Default: developers with more than three models; unlisted developers hidden. All records remain in the evidence table.
Select a point or choose a model to inspect its score and uncertainty.
ECI is an index, not a percentage or a direct measure of reliable work. The step line is the best score by release date within the selected group, not a forecast. No future capability is extrapolated.
Source: Epoch AI · publication date unlisted ↗ · Epoch AI · publication date unlisted ↗. Retrieved 2026-09-09. Data © Epoch AI and contributors, reused with attribution. 90% confidence intervals come from the same download; two calibration anchors have no supplied interval.
Country comes from Epoch’s “Country (of organization)” field; it is not the training location or the nationality of every contributor. Unknown values remain unlisted. Developer labels retain Epoch’s attribution, including collaborations. Historical scores can change when Epoch refits its index. This page uses one complete snapshot rather than combining vintages. Benchmark coverage and developer-reported results can affect comparability; overlapping intervals do not establish a reliable ranking. Open weights do not necessarily mean unrestricted open source.
| Model | Developer | Country of organization | Release date | ECI | 90% interval | Availability |
|---|---|---|---|---|---|---|
| GPT-6 Astra | OpenAI | United States of America | 2026-09-03 | 166.57 | 163.34–171.7 | Closed weights |
| Claude Fable 5.1 | Anthropic | United States of America | 2026-09-01 | 164.24 | 161.3–167.85 | Closed weights |
| Claude Fable 5 | Anthropic | United States of America | 2026-06-09 | 163.41 | 160.68–167.1 | Closed weights |
| Claude Opus 5 | Anthropic | United States of America | 2026-07-24 | 162.59 | 160.09–165.69 | Closed weights |
| GPT-5.5 Pro | OpenAI | United States of America | 2026-04-23 | 162.26 | 159.3–165.37 | Closed weights |
| GPT-5.6 Sol | OpenAI | United States of America | 2026-07-09 | 161.76 | 159.53–164.88 | Closed weights |
| GPT-5.6 Terra | OpenAI | United States of America | 2026-07-09 | 159.15 | 156.64–161.84 | Closed weights |
| GPT-5.5 | OpenAI | United States of America | 2026-04-23 | 159.14 | 156.91–161.87 | Closed weights |
| GPT-5.4 Pro | OpenAI | United States of America | 2026-03-05 | 158.93 | 156.35–161.87 | Closed weights |
| Claude Opus 4.8 | Anthropic | United States of America | 2026-05-28 | 158.32 | 156.33–160.46 | Closed weights |
| Kimi K3 | Moonshot | China | 2026-07-16 | 157.64 | 155.01–160.58 | Open weights |
| Gemini 3.7 Flash | Google DeepMind | United States of America | 2026-08-13 | 157.44 | 155.52–160.08 | Closed weights |
| GPT-5.4 | OpenAI | United States of America | 2026-03-05 | 156.86 | 155.05–158.79 | Closed weights |
| Qwen 3.8 Max | Alibaba | China | 2026-08-02 | 156.62 | 154.45–158.45 | Closed weights |
| GPT-5.3 Codex | OpenAI | United States of America | 2026-02-05 | 156.59 | 154.13–159.41 | Closed weights |
| Claude Opus 4.7 | Anthropic | United States of America | 2026-04-16 | 156.34 | 154.45–158.34 | Closed weights |
| Grok 4.6 | xAI | United States of America | 2026-08-12 | 156.34 | 154.57–158.47 | Closed weights |
| GPT-5.6 Luna | OpenAI | United States of America | 2026-07-09 | 156.31 | 153.75–158.49 | Closed weights |
| Claude Sonnet 5 | Anthropic | United States of America | 2026-06-30 | 156.23 | 153.39–158.5 | Closed weights |
| Muse Spark 1.2 | Meta AI | United States of America | 2026-08-05 | 155.49 | 153.76–158.7 | Closed weights |
| DeepSeek V4 Pro 0813 | DeepSeek | China | 2026-08-13 | 155.44 | 153.65–157.54 | Open weights |
| Claude Opus 4.6 | Anthropic | United States of America | 2026-02-05 | 155.34 | 153.51–157.14 | Closed weights |
| GPT-5.2 Pro | OpenAI | United States of America | 2025-12-11 | 155.34 | 153.04–157.88 | Closed weights |
| GLM-5.3 | Z.ai (Zhipu AI) | China | 2026-08-14 | 155.25 | 153.15–157.51 | Closed weights |
| Gemini 3.1 Pro | Google DeepMind | United States of America | 2026-02-19 | 155 | 152.77–157.56 | Closed weights |
| Gemini 3.5 Flash | Google DeepMind | United States of America | 2026-05-19 | 154.69 | 152.76–156.78 | Closed weights |
| Muse Spark 1.1 | Meta AI | United States of America | 2026-07-09 | 154.64 | 152.26–157.96 | Closed weights |
| DeepSeek V4 Flash 0731 | DeepSeek | China | 2026-07-31 | 154.49 | 152.05–156.45 | Open weights |
| Gemini 3.6 Flash | Google DeepMind | United States of America | 2026-07-21 | 154.32 | 152.6–156.05 | Closed weights |
| Grok 4.5 | xAI | United States of America | 2026-07-08 | 153.92 | 152.32–155.71 | Closed weights |
| Qwen3.7-Max | Alibaba | China | 2026-05-19 | 153.72 | 151.87–155.95 | Closed weights |
| GPT-5.2 | OpenAI | United States of America | 2025-12-11 | 153.49 | 151.77–155.17 | Closed weights |
| Gemini 3 Pro | Google DeepMind | United States of America | 2025-11-18 | 153.01 | 151.16–154.96 | Closed weights |
| Claude Sonnet 4.6 | Anthropic | United States of America | 2026-02-17 | 152.29 | 149.78–154.35 | Closed weights |
| Muse Spark | Meta AI | United States of America | 2026-04-08 | 152.13 | 150.18–155.41 | Closed weights |
| Grok 4.20 | xAI | United States of America | 2026-02-17 | 152.03 | 149.08–154.45 | Closed weights |
| GLM-5.2 | Z.ai (Zhipu AI) | China | 2026-06-16 | 151.88 | 150.11–153.88 | Open weights |
| Gemini 3 Flash | Google DeepMind | United States of America | 2025-12-17 | 151.84 | 150.5–153.45 | Closed weights |
| GLM-5.3-Flash | Z.ai (Zhipu AI) | China | 2026-08-20 | 151.45 | 148.96–153.42 | Open weights |
| Kimi K2.6 | Moonshot | China | 2026-04-20 | 151.01 | 149.07–152.65 | Open weights |
| GPT-5 Pro | OpenAI | United States of America | 2025-10-07 | 150.28 | 148.77–152.52 | Closed weights |
| Inkling-Small | Thinking Machines | United States of America | 2026-07-15 | 150.17 | 147.44–151.96 | Open weights |
| Claude Opus 4.5 | Anthropic | United States of America | 2025-11-24 | 150.11 | 148.0–152.85 | Closed weights |
| Kimi K2.7 Code | Moonshot | China | 2026-06-12 | 150.1 | 148.39–151.76 | Open weights |
| GPT-5 | OpenAI | United States of America | 2025-08-07 | 150 | Anchor; interval not supplied | Closed weights |
| GLM-5.1 | Z.ai (Zhipu AI) | China | 2026-04-07 | 149.73 | 147.65–151.48 | Open weights |
| GPT-5.1 | OpenAI | United States of America | 2025-11-13 | 149.65 | 148.34–151.05 | Closed weights |
| Qwen 3.6 Max (Preview) | Alibaba | China | 2026-04-20 | 149.3 | 147.55–152.27 | Other |
| Grok 4.3 Beta | xAI | United States of America | 2026-04-17 | 149.18 | 147.59–150.59 | Closed weights |
| DeepSeek-V4-Pro | DeepSeek | China | 2026-04-24 | 149.1 | 147.29–150.44 | Open weights |
| GPT-5.4 Mini | OpenAI | United States of America | 2026-03-17 | 148.98 | 147.08–150.7 | Closed weights |
| Inkling | Thinking Machines | United States of America | 2026-07-15 | 148.64 | 145.93–150.5 | Open weights |
| Kimi K2.5 | Moonshot | China | 2026-01-27 | 148.03 | 146.24–149.09 | Open weights |
| Qwen 3.6 Plus | Alibaba | China | 2026-03-31 | 147.69 | 145.8–149.26 | Closed weights |
| o3-pro | OpenAI | United States of America | 2025-06-10 | 147.45 | 145.86–150.15 | Closed weights |
| Qwen3.7-Plus | Alibaba | China | 2026-06-02 | 147.42 | 145.83–148.78 | Other |
| Qwen3.5 397B-A17B | Alibaba | China | 2026-02-13 | 146.98 | 145.27–148.83 | Open weights |
| o3 | OpenAI | United States of America | 2025-04-16 | 146.91 | 144.99–148.44 | Closed weights |
| Claude Sonnet 4.5 | Anthropic | United States of America | 2025-09-29 | 146.85 | 145.32–148.21 | Closed weights |
| Qwen 3.5 Plus (hosted 397B-A17B) | Alibaba | China | 2026-02-16 | 146.75 | 144.6–147.94 | Other |
| MiniMax-M2.5 | MiniMax | China | 2026-02-12 | 146.52 | 138.5–147.57 | Open weights |
| Qwen3.6 27B | Alibaba | China | 2026-04-22 | 146.51 | 144.53–147.8 | Open weights |
| MiniMax-M3 | MiniMax | China | 2026-06-01 | 146.5 | 142.31–149.19 | Open weights |
| Grok 4 | xAI | United States of America | 2025-07-09 | 146.45 | 144.39–148.14 | Closed weights |
| Nemotron 3 Ultra | NVIDIA | United States of America | 2026-06-04 | 146.29 | 143.85–148.27 | Open weights |
| DeepSeek-V3.2 | DeepSeek | China | 2025-12-01 | 146.2 | 144.34–147.39 | Open weights |
| DeepSeek-V4-Flash | DeepSeek | China | 2026-04-24 | 146.12 | 143.67–147.87 | Open weights |
| GPT-5.4 Nano | OpenAI | United States of America | 2026-03-17 | 145.87 | 143.46–147.52 | Closed weights |
| GLM-5 | Z.ai (Zhipu AI) | China | 2026-02-11 | 145.87 | 143.94–147.8 | Open weights |
| MiniMax-M2.7 | MiniMax | China | 2026-03-18 | 145.81 | 138.38–147.74 | Open weights |
| Kimi K2 Thinking | Moonshot | China | 2025-11-06 | 145.77 | 143.31–147.12 | Open weights |
| o4-mini | OpenAI | United States of America | 2025-04-16 | 145.66 | 143.29–147.27 | Closed weights |
| GPT-5 mini | OpenAI | United States of America | 2025-08-07 | 145.54 | 143.51–147.11 | Closed weights |
| Gemini 2.5 Pro (Jun 2025) | Google DeepMind | United States of America | 2025-06-05 | 145.27 | 143.59–146.71 | Closed weights |
| Gemini 3.5 Flash-Lite | Google DeepMind | United States of America | 2026-07-21 | 145.14 | 142.4–146.7 | Closed weights |
| DeepSeek-V3.2-Exp | DeepSeek | China | 2025-09-29 | 145.03 | 142.53–147.0 | Open weights |
| Qwen3.7 Flash | Alibaba | China | 2026-07-27 | 144.64 | 142.69–147.52 | Closed weights |
| Gemini 3.1 Flash-Lite | United States of America | 2026-03-03 | 144.5 | 142.72–145.98 | Closed weights | |
| Grok 4 Fast | xAI | United States of America | 2025-09-19 | 144.22 | 141.74–146.28 | Closed weights |
| Gemini 2.5 Pro (Mar 2025) | Google DeepMind | United States of America | 2025-03-31 | 144.17 | 141.94–146.84 | Closed weights |
| Claude Opus 4.1 | Anthropic | United States of America | 2025-08-05 | 144.12 | 141.76–145.99 | Closed weights |
| Qwen 3.5 Flash (hosted 35B-A3B) | Alibaba | China | 2026-02-25 | 144.01 | 141.92–145.67 | Other |
| Qwen3-235B-A22B-Thinking (Jul 2025) | Alibaba | China | 2025-07-25 | 143.88 | 141.77–145.36 | Open weights |
| Qwen 3.6 35B-A3B | Alibaba | China | 2026-04-14 | 143.87 | 141.31–146.0 | Open weights |
| GLM-4.7 | Z.ai (Zhipu AI) | China | 2025-12-22 | 143.47 | 141.16–145.49 | Open weights |
| Qwen 3.6 Flash | Alibaba | China | 2026-04-27 | 143.28 | 140.98–144.78 | Other |
| Gemini 2.5 Flash (Sep 2025) | Google DeepMind | United States of America | 2025-09-25 | 143 | 139.77–146.62 | Closed weights |
| Claude Opus 4 | Anthropic | United States of America | 2025-05-22 | 142.68 | 140.58–144.11 | Closed weights |
| Gemma 4 31B IT | Google DeepMind | United States of America | 2026-04-02 | 142.68 | 140.28–144.75 | Open weights |
| Qwen3.5-35B-A3B | Alibaba | China | 2026-02-24 | 142.54 | 140.22–144.87 | Open weights |
| GPT-5.5 Instant | OpenAI | United States of America | 2026-05-05 | 142.52 | 139.99–145.21 | Closed weights |
| Gemini 2.5 Pro (May 2025) | Google DeepMind | United States of America | 2025-05-06 | 142.48 | 139.15–145.14 | Closed weights |
| Qwen3-Max | Alibaba | China | 2025-09-24 | 142.44 | 140.12–144.73 | Closed weights |
| Claude Haiku 4.5 | Anthropic | United States of America | 2025-10-15 | 142.39 | 139.47–144.03 | Closed weights |
| Gemma 4 26B A4B | Google DeepMind | United States of America | 2026-04-02 | 141.87 | 138.55–143.58 | Open weights |
| o1 | OpenAI | United States of America | 2024-12-17 | 141.87 | 139.83–143.14 | Closed weights |
| Claude Sonnet 4 | Anthropic | United States of America | 2025-05-22 | 141.69 | 138.85–143.08 | Closed weights |
| Gemini 2.5 Flash (May 2025) | Google DeepMind | United States of America | 2025-05-20 | 141.55 | 139.51–142.98 | Closed weights |
| DeepSeek-R1 (May 2025) | DeepSeek | China | 2025-05-28 | 141.3 | 139.17–142.91 | Open weights |
| Mistral Medium 3.5 | Mistral AI | France | 2026-04-28 | 141.29 | 138.78–143.1 | Open weights |
| Claude 3.7 Sonnet | Anthropic | United States of America | 2025-02-24 | 141.17 | 139.01–142.68 | Closed weights |
| GLM-4.6 | Z.ai (Zhipu AI),Tsinghua University | China | 2025-09-30 | 140.79 | 134.21–142.64 | Open weights |
| Gemini 2.5 Flash (Jun 2025) | Google DeepMind | United States of America | 2025-06-17 | 140.53 | 138.18–141.98 | Closed weights |
| o3-mini | OpenAI | United States of America | 2025-01-31 | 140.36 | 137.47–141.8 | Closed weights |
| Grok-3 mini | xAI | United States of America | 2025-06-24 | 140.36 | 137.67–142.02 | Closed weights |
| Kimi K2 (Jul 2025) | Moonshot | China | 2025-07-12 | 140.12 | 136.65–142.03 | Open weights |
| gpt-oss-120b | OpenAI | United States of America | 2025-08-05 | 140.11 | 135.39–142.42 | Open weights |
| Gemini 2.5 Flash (Apr 2025) | Google DeepMind | United States of America | 2025-04-17 | 139.98 | 136.37–142.05 | Closed weights |
| DeepSeek-V3.1 | DeepSeek | China | 2025-08-21 | 139.93 | 135.79–143.19 | Open weights |
| Qwen3-30B-A3B-Thinking (Jul 2025) | Alibaba | China | 2025-07-30 | 139.65 | 135.15–141.06 | Open weights |
| Qwen3.5-9B | Alibaba | China | 2026-02-24 | 139.45 | 136.59–141.07 | Open weights |
| GPT-5 nano | OpenAI | United States of America | 2025-08-07 | 139.4 | 134.43–141.52 | Closed weights |
| Qwen3-235B-A22B | Alibaba | China | 2025-04-28 | 139.36 | 135.59–140.99 | Open weights |
| DeepSeek-R1 | DeepSeek | China | 2025-01-20 | 138.98 | 136.62–140.38 | Open weights |
| Qwen3-235B-A22B-Instruct (Jul 2025) | Alibaba | China | 2025-07-25 | 138.93 | 136.19–140.82 | Open weights |
| Qwen3-32B | Alibaba | China | 2025-04-29 | 138.52 | 135.14–140.32 | Open weights |
| Grok 3 | xAI | United States of America | 2025-04-09 | 138.3 | 135.61–139.8 | Closed weights |
| Qwen3-14B | Alibaba | China | 2025-04-29 | 138.25 | 133.68–139.84 | Open weights |
| gpt-oss-20b | OpenAI | United States of America | 2025-08-05 | 137.81 | 132.89–139.56 | Open weights |
| QwQ-32B | Alibaba | China | 2025-03-05 | 137.6 | 133.05–141.72 | Open weights |
| DeepSeek-R1-Distill-Qwen-32B | DeepSeek | China | 2025-01-20 | 137.43 | 126.69–138.7 | Open weights |
| Qwen3-30B-A3B-Instruct (Jul 2025) | Alibaba | China | 2025-07-29 | 137.43 | 132.32–139.54 | Open weights |
| GPT-4.1 | OpenAI | United States of America | 2025-04-14 | 136.8 | 134.21–138.54 | Closed weights |
| GPT-4.5 | OpenAI | United States of America | 2025-02-27 | 136.75 | 133.97–138.6 | Closed weights |
| Qwen3-30B-A3B | Alibaba | China | 2025-04-29 | 136.19 | 130.82–138.76 | Open weights |
| Qwen3-8B | Alibaba | China | 2025-04-28 | 136.18 | 129.58–138.07 | Open weights |
| DeepSeek-V3 (Mar 2025) | DeepSeek | China | 2025-03-24 | 135.95 | 132.46–138.04 | Open weights |
| o1-mini | OpenAI | United States of America | 2024-09-12 | 135.83 | 132.4–137.38 | Closed weights |
| DeepSeek-R1-Distill-Qwen-14B | DeepSeek | China | 2025-01-20 | 135.43 | 127.36–138.05 | Open weights |
| Gemini 2.0 Flash Thinking (Jan 2025) | Google DeepMind,Google | United States of America | 2025-01-21 | 135.37 | 130.11–137.99 | Closed weights |
| Gemini 2.0 Pro | Google DeepMind | United States of America | 2025-02-05 | 135.06 | 131.28–136.85 | Closed weights |
| GPT-4.1 mini | OpenAI | United States of America | 2025-04-14 | 135.02 | 131.5–136.71 | Closed weights |
| o1-preview | OpenAI | United States of America | 2024-09-12 | 134.79 | 131.47–138.65 | Closed weights |
| Gemini 2.0 Flash (Dec 2024) | Google DeepMind,Google | United States of America | 2024-12-11 | 134.71 | 124.35–136.59 | Closed weights |
| Gemini 2.0 Flash (Feb 2025) | Google DeepMind,Google | United States of America | 2025-02-05 | 134.69 | 131.46–136.36 | Closed weights |
| Mistral Medium 3 | Mistral AI | France | 2025-05-07 | 134.08 | 130.69–135.57 | Closed weights |
| Gemini 2.5 Flash-Lite (Jun 2025) | Google DeepMind | United States of America | 2025-06-17 | 133.93 | 130.4–136.41 | Closed weights |
| Claude 3.5 Sonnet (October 2024) | Anthropic | United States of America | 2024-10-22 | 133.55 | 128.63–137.61 | Closed weights |
| Magistral Small 1.0 | Mistral AI | France | 2025-06-10 | 133.19 | 129.42–134.52 | Open weights |
| Qwen2.5-Max | Alibaba | China | 2025-01-25 | 132.54 | 128.45–135.82 | Closed weights |
| DeepSeek-V3 | DeepSeek | China | 2024-12-26 | 132.35 | 128.05–135.89 | Open weights |
| Llama 4 Maverick | Meta AI | United States of America | 2025-04-06 | 132.2 | 128.31–134.35 | Open weights |
| Mistral Small 3.2 | Mistral AI | France | 2025-06-20 | 131.74 | 127.19–133.73 | Open weights |
| Gemini 1.5 Pro (Sept 2024) | Google DeepMind | United States of America | 2024-09-24 | 131.73 | 128.35–133.24 | Closed weights |
| Magistral Small 1.2 | Mistral AI | France | 2025-09-18 | 131.41 | 126.27–133.53 | Open weights |
| Grok-2 (Dec 2024) | xAI | United States of America | 2024-12-12 | 130.48 | 126.47–132.12 | Closed weights |
| Phi-4 | Microsoft Research | United States of America | 2024-12-12 | 130.42 | 125.5–132.1 | Open weights |
| Gemma 3 27B | Google DeepMind | United States of America | 2025-03-12 | 130.04 | 124.64–132.17 | Open weights |
| Claude 3.5 Sonnet | Anthropic | United States of America | 2024-06-20 | 130 | Anchor; interval not supplied | Closed weights |
| Llama 4 Scout | Meta AI | United States of America | 2025-04-05 | 129.64 | 124.84–131.45 | Open weights |
| GPT-4.1 nano | OpenAI | United States of America | 2025-04-14 | 129.62 | 124.04–132.0 | Closed weights |
| Gemini 1.5 Flash (Sep 2024) | Google DeepMind | United States of America | 2024-09-24 | 129.36 | 124.03–131.09 | Closed weights |
| Qwen2.5-72B | Alibaba | China | 2024-09-19 | 129 | 123.98–130.7 | Open weights |
| GPT-4o (May 2024) | OpenAI | United States of America | 2024-05-13 | 128.97 | 124.43–131.59 | Closed weights |
| GPT-4o (Nov 2024) | OpenAI | United States of America | 2024-11-20 | 128.81 | 124.92–130.93 | Closed weights |
| GPT-4o (Aug 2024) | OpenAI | United States of America | 2024-08-06 | 128.77 | 124.39–130.6 | Closed weights |
| Llama 3.1-405B | Meta AI | United States of America | 2024-07-23 | 128.75 | 124.39–130.57 | Open weights |
| Qwen2.5-32B | Alibaba | China | 2024-09-17 | 128.52 | 123.16–130.09 | Open weights |
| Mistral Large 2 (Nov 2024) | Mistral AI | France | 2024-11-18 | 128.52 | 124.19–130.55 | Open weights |
| Mistral Large 2 (Jul 2024) | Mistral AI | France | 2024-07-24 | 127.54 | 123.07–129.64 | Open weights |
| Mistral Small 3.1 | Mistral AI | France | 2025-03-17 | 127.48 | 122.92–129.28 | Open weights |
| Llama 3.3 70B | Meta AI | United States of America | 2024-12-06 | 127.32 | 122.68–129.67 | Open weights |
| GPT-4 Turbo (Apr 2024) | OpenAI | United States of America | 2024-04-09 | 127.25 | 122.46–128.93 | Closed weights |
| Claude 3.5 Haiku | Anthropic | United States of America | 2024-10-22 | 127.15 | 121.02–129.64 | Closed weights |
| Mistral Small 3 | Mistral AI | France | 2025-01-30 | 127.07 | 122.41–129.01 | Open weights |
| Claude 3 Opus | Anthropic | United States of America | 2024-02-29 | 126.91 | 121.74–130.11 | Closed weights |
| Gemini 1.5 Pro (May 2024) | Google DeepMind | United States of America | 2024-05-14 | 126.9 | 122.52–129.03 | Closed weights |
| GPT-4o mini | OpenAI | United States of America | 2024-07-18 | 126.56 | 121.2–128.53 | Closed weights |
| GPT-4 Turbo (Nov 2023) | OpenAI | United States of America | 2024-01-25 | 126.45 | 121.78–128.95 | Closed weights |
| Llama 3.1-70B | Meta AI | United States of America | 2024-07-23 | 125.91 | 121.1–128.39 | Open weights |
| GPT-4 (Mar 2023) | OpenAI | United States of America | 2023-03-14 | 125.88 | 120.22–130.04 | Closed weights |
| Llama 3.2 90B | Meta AI | United States of America | 2024-09-24 | 125.5 | 120.49–127.47 | Open weights |
| Qwen2-72B | Alibaba | China | 2024-06-07 | 125.27 | 119.75–126.9 | Open weights |
| DeepSeek-V2 (MoE-236B, May 2024) | DeepSeek | China | 2024-05-07 | 124.77 | 118.57–127.57 | Open weights |
| Amazon Nova Pro | Amazon | United States of America | 2024-12-03 | 123.78 | 107.53–126.62 | Closed weights |
| Gemma 3 12B | Google DeepMind | United States of America | 2025-03-12 | 123.46 | 116.21–128.73 | Open weights |
| GPT-4 (Jun 2023) | OpenAI | United States of America | 2023-06-13 | 123.1 | 117.72–126.7 | Closed weights |
| Llama 3-70B | Meta AI | United States of America | 2024-04-18 | 122.91 | 117.38–125.73 | Open weights |
| Gemini 1.5 Flash (May 2024) | Google DeepMind | United States of America | 2024-05-23 | 122.57 | 116.81–124.7 | Closed weights |
| Gemma 2 27B | Google DeepMind | United States of America | 2024-06-24 | 122.05 | 115.94–124.18 | Open weights |
| Mixtral 8x22B | Mistral AI | France | 2024-04-17 | 122 | 115.86–124.39 | Open weights |
| Mistral Large | Mistral AI | France | 2024-02-26 | 121.99 | 114.74–124.93 | Closed weights |
| phi-3-small 7.4B | Microsoft | United States of America | 2024-04-23 | 121.8 | 114.95–124.24 | Open weights |
| phi-3-medium 14B | Microsoft | United States of America | 2024-04-23 | 121.19 | 115.17–123.87 | Open weights |
| Claude 3 Sonnet | Anthropic | United States of America | 2024-02-29 | 120.68 | 113.89–123.5 | Closed weights |
| Claude Instant | Anthropic | United States of America | 2023-08-09 | 120.21 | 113.43–122.93 | Closed weights |
| Claude 2 | Anthropic | United States of America | 2023-07-11 | 120.09 | 114.21–123.8 | Closed weights |
| Gemma 2 9B | Google DeepMind | United States of America | 2024-06-24 | 119.8 | 112.44–122.31 | Open weights |
| Qwen2.5-Coder-32B | Alibaba | China | 2024-09-18 | 119.43 | 110.31–125.07 | Open weights |
| Command R+ | Cohere,Cohere Labs (formerly Cohere for AI) | Canada | 2024-08-30 | 119.25 | 110.28–125.06 | Open weights |
| Claude 2.1 | Anthropic | United States of America | 2023-11-21 | 119.22 | 113.07–121.92 | Closed weights |
| Mistral NeMo | Mistral AI | France | 2024-07-18 | 118.65 | 111.96–121.85 | Open weights |
| GPT-3.5 Turbo (Nov 2023) | OpenAI | United States of America | 2023-11-06 | 118.5 | 108.73–121.55 | Closed weights |
| Qwen2.5-7B | Alibaba | China | 2024-09-19 | 118.45 | 111.07–121.28 | Open weights |
| Mixtral 8x7B | Mistral AI | France | 2023-12-11 | 118.4 | 111.79–121.15 | Open weights |
| Claude 3 Haiku | Anthropic | United States of America | 2024-03-07 | 118.3 | 111.75–121.5 | Closed weights |
| Ministral 3B | Mistral AI | France | 2024-10-16 | 118.06 | 107.01–121.85 | Closed weights |
| Yi-34B | 01.AI | China | 2023-11-02 | 117.31 | 108.1–120.22 | Open weights |
| phi-3-mini 3.8B | Microsoft | United States of America | 2024-04-23 | 117.27 | 109.44–120.54 | Open weights |
| Stable Beluga 2 | Stability AI | United Kingdom | 2023-07-20 | 117.02 | 110.39–120.14 | Open weights |
| Gemini 1.0 Pro | Google DeepMind | United States of America | 2023-12-13 | 116.97 | 110.31–119.74 | Closed weights |
| Llama 3.1-8B | Meta AI | United States of America | 2024-07-23 | 116.51 | 105.95–121.71 | Open weights |
| Llama 3-8B | Meta AI | United States of America | 2024-04-18 | 116.36 | 108.36–119.1 | Open weights |
| Qwen2.5-Coder-14B | Unknown | Unknown | 2024-09-18 | 116.3 | 106.61–122.34 | Other |
| Gemma 3 4B | Google DeepMind | United States of America | 2025-03-12 | 115.97 | 97.45–123.39 | Open weights |
| GPT-3.5 Turbo (Jan 2024) | OpenAI | United States of America | 2024-01-25 | 115.62 | 107.2–119.14 | Closed weights |
| PaLM 2-L | Unknown | Unknown | 2023-05-17 | 114.9 | 105.47–127.71 | Other |
| Llama 2-70B | Meta AI | United States of America | 2023-07-18 | 113.65 | 106.09–117.31 | Open weights |
| GPT-3.5 Turbo (Jun 2023) | OpenAI | United States of America | 2023-06-13 | 113.21 | 105.97–117.89 | Closed weights |
| Qwen2.5-Coder (7B) | Alibaba | China | 2024-09-18 | 113 | 101.97–118.44 | Open weights |
| Qwen-14B | Alibaba | China | 2023-09-24 | 112.88 | 105.14–117.16 | Open weights |
| Mistral 7B v0.1 | Mistral AI | France | 2023-09-27 | 112.05 | 103.93–116.19 | Open weights |
| Falcon-180B | Technology Innovation Institute | United Arab Emirates | 2023-09-06 | 111.97 | 104.16–118.52 | Open weights |
| internlm-20b | Unknown | Unknown | 2023-09-18 | 111.92 | 103.27–117.01 | Other |
| Gemma 7B | Google DeepMind | United States of America | 2024-02-21 | 111.82 | 103.19–116.2 | Open weights |
| DeepSeek LLM 67B | DeepSeek | China | 2023-11-29 | 110.55 | 93.83–117.03 | Open weights |
| LLaMA-65B | Meta AI | United States of America | 2023-02-24 | 109.98 | 102.15–114.27 | Open weights |
| Falcon 2 11B | Technology Innovation Institute | United Arab Emirates | 2024-05-09 | 109.36 | 101.39–115.43 | Open weights |
| Mistral 7B v0.3 | Mistral AI | France | 2024-05-27 | 108.81 | 98.65–112.44 | Open weights |
| DeepSeek-Coder-V2-Lite-Base | Unknown | Unknown | 2024-06-13 | 108.68 | 99.7–113.87 | Other |
| PaLM 2-M | Unknown | Unknown | 2023-05-17 | 108.08 | 99.26–115.99 | Other |
| Phi-2 | Microsoft | United States of America | 2023-12-12 | 107.73 | 88.55–113.35 | Open weights |
| Qwen2.5-Coder-3B | Unknown | Unknown | 2024-09-18 | 107.52 | 94.3–115.58 | Other |
| Nemotron-4 15B | NVIDIA | United States of America | 2024-02-26 | 107.49 | 99.12–112.96 | Closed weights |
| Yi-9B | Unknown | Unknown | 2024-03-01 | 107.39 | 98.62–114.57 | Other |
| LLaMA-33B | Meta AI | United States of America | 2023-02-24 | 107.18 | 99.0–112.24 | Open weights |
| Qwen-7B | Alibaba | China | 2023-09-28 | 106.59 | 95.55–112.2 | Open weights |
| Llama 2-13B | Meta AI | United States of America | 2023-07-18 | 105.93 | 96.7–111.03 | Open weights |
| PaLM 2-S | Unknown | Unknown | 2023-05-17 | 105.92 | 97.66–113.71 | Other |
| Llama 2-34B | Meta AI | United States of America | 2023-07-18 | 104.94 | 95.48–110.32 | Closed weights |
| StarCoder 2 15B | Hugging Face,ServiceNow,NVIDIA,BigCode | United States of America | 2024-02-20 | 104.74 | 92.95–111.43 | Open weights |
| Yi 6B | 01.AI | China | 2023-11-22 | 104.51 | 95.54–110.28 | Open weights |
| Falcon-40B | Technology Innovation Institute | United Arab Emirates | 2023-05-25 | 104.18 | 94.62–109.74 | Open weights |
| Baichuan2-13B | Baichuan | China | 2023-09-06 | 102.91 | 91.79–110.02 | Open weights |
| Qwen2.5-Coder (1.5B) | Alibaba | China | 2024-09-18 | 102.71 | 88.02–110.25 | Open weights |
| internlm-7b | Unknown | Unknown | 2023-07-05 | 102.58 | 91.92–109.22 | Other |
| Llama 3.2 1B | Meta AI | United States of America | 2024-09-24 | 102.43 | 91.24–109.91 | Open weights |
| INTELLECT-1 | Prime Intellect,Hugging Face,Arcee AI | United States of America | 2024-11-29 | 100.55 | 90.42–106.13 | Other |
| MPT-30B | MosaicML | United States of America | 2023-06-22 | 100.34 | 90.51–106.09 | Open weights |
| LLaMA-13B | Meta AI | United States of America | 2023-02-24 | 100.26 | 89.93–106.41 | Open weights |
| Llama 2-7B | Meta AI | United States of America | 2023-07-18 | 98.72 | 88.42–105.23 | Open weights |
| chatglm2-6b | Unknown | Unknown | 2023-06-24 | 98.6 | 83.28–105.31 | Other |
| LLaMA-7B | Meta AI | United States of America | 2023-02-24 | 96.32 | 85.32–103.57 | Open weights |
| Baichuan 2-7B | Baichuan | China | 2023-09-20 | 95.95 | 85.1–104.4 | Open weights |
| DeepSeek Coder 33B | DeepSeek,Peking University | China | 2023-11-02 | 95.94 | 83.94–102.96 | Open weights |
| Falcon-7B | Technology Innovation Institute | United Arab Emirates | 2023-04-24 | 94.7 | 83.52–101.96 | Open weights |
| CodeQwen1.5-7B | Unknown | Unknown | 2024-04-15 | 94.46 | 81.41–102.8 | Other |
| vicuna-13b-v1.1 | Unknown | Unknown | 2023-04-12 | 94.26 | 80.79–102.31 | Other |
| MPT-7B | MosaicML | United States of America | 2023-05-05 | 94.19 | 81.83–101.65 | Open weights |
| Gemma 2B | Google DeepMind | United States of America | 2024-02-21 | 93.8 | 82.78–100.32 | Open weights |
| StarCoder 2 7B | Hugging Face,ServiceNow,NVIDIA,BigCode | United States of America | 2024-02-20 | 93.15 | 77.74–101.69 | Open weights |
| XGen-7B | Salesforce | United States of America | 2023-06-27 | 92.98 | 82.08–100.15 | Open weights |
| Qwen-1_8B | Unknown | Unknown | 2023-11-30 | 92.51 | 70.92–100.95 | Other |
| open_llama_7b | Unknown | Unknown | 2023-06-07 | 91.21 | 79.11–99.45 | Other |
| Phi-1.5 | Microsoft | United States of America | 2023-09-11 | 91.08 | 68.51–100.38 | Open weights |
| Baichuan1-7B | Baichuan | China | 2023-06-01 | 90.01 | 77.3–98.98 | Open weights |
| RedPajama-INCITE-7B-Base | Unknown | Unknown | 2023-05-04 | 89.96 | 76.67–98.05 | Other |
| Dolly 2.0-12b | Databricks | United States of America | 2023-04-11 | 89.2 | 75.7–97.11 | Open weights |
| DeepSeek Coder 6.7B | DeepSeek,Peking University | China | 2023-11-02 | 89.15 | 75.47–97.29 | Open weights |
| StarCoder 2 3B | Hugging Face,ServiceNow,NVIDIA,BigCode | United States of America | 2024-02-22 | 88.28 | 73.61–97.17 | Open weights |
| Qwen2.5-Coder-0.5B | Unknown | Unknown | 2024-09-18 | 87.66 | 62.09–99.7 | Other |
| Cerebras-GPT-13B | Cerebras Systems | United States of America | 2023-03-20 | 82.68 | 67.57–91.55 | Open weights |
| DeepSeek Coder 1.3B | DeepSeek,Peking University | China | 2023-11-02 | 62.73 | 46.12–76.59 | Open weights |
| stablelm-tuned-alpha-7b | Unknown | Unknown | 2023-04-19 | 54.83 | 31.39–72.75 | Other |
Download reviewed model snapshot ↗ · Dataset licensing & original authors ↗