| # | Model | Score | Relative | Setting | Released | Run | Evidence |
|---|---|---|---|---|---|---|---|
| 1 | Llama 2-70B Open source Meta AI | 87.6% | — | 18 Jul 2023 | — | Source ↗ | |
| 2 | Claude 2 Anthropic | 87.5% | — | 11 Jul 2023 | — | Source ↗ | |
| 3 | Claude 1.3 Anthropic | 86.7% | — | 18 Apr 2023 | — | Source ↗ | |
| 4 | PaLM 2-L | 86.1% | — | 17 May 2023 | — | Source ↗ | |
| 5 | LLaMA-65B Open source Meta AI | 86.0% | — | 24 Feb 2023 | — | Source ↗ | |
| 6 | GPT-3.5 Turbo (Nov 2023) OpenAI | 85.8% | — | 6 Nov 2023 | — | Source ↗ | |
| 7 | GPT-4 (Jun 2023) OpenAI | 84.8% | — | 13 Jun 2023 | — | Source ↗ | |
| 8 | Llama 2-34B Meta AI | 84.6% | — | 18 Jul 2023 | — | Source ↗ | |
| 9 | LLaMA-33B Open source Meta AI | 83.8% | — | 24 Feb 2023 | — | Source ↗ | |
| 10 | DeepSeek V3 Open source DeepSeek | 82.9% | — | 26 Dec 2024 | — | Source ↗ | |
| 11 | Llama 3.1-405B Open source Meta AI | 82.7% | — | 23 Jul 2024 | — | Source ↗ | |
| 12 | Mixtral 8x7B Open source Mistral AI | 82.2% | — | 11 Dec 2023 | — | Source ↗ | |
| 13 | PaLM 2-M | 81.7% | — | 17 May 2023 | — | Source ↗ | |
| 14 | DeepSeek-V2 (MoE-236B, May 2024) Open source DeepSeek | 80.0% | — | 7 May 2024 | — | Source ↗ | |
| 15 | Falcon-40B Open source Technology Innovation Institute | 79.9% | — | 25 May 2023 | — | Source ↗ | |
| 16 | Llama 2-13B Open source Meta AI | 79.6% | — | 18 Jul 2023 | — | Source ↗ | |
| 17 | Claude Instant Anthropic | 78.9% | — | 9 Aug 2023 | — | Source ↗ | |
| 18 | LLaMA-13B Open source Meta AI | 77.9% | — | 24 Feb 2023 | — | Source ↗ | |
| 19 | Mistral 7B v0.1 Open source Mistral AI | 75.2% | — | 27 Sep 2023 | — | Source ↗ | |
| 20 | PaLM 2-S | 75.2% | — | 17 May 2023 | — | Source ↗ | |
| 21 | phi-3-medium 14B Open source Microsoft | 73.9% | — | 23 Apr 2024 | — | Source ↗ | |
| 22 | Llama 2-7B Open source Meta AI | 73.7% | — | 18 Jul 2023 | — | Source ↗ | |
| 23 | MPT-30B Open source MosaicML | 73.6% | — | 22 Jun 2023 | — | Source ↗ | |
| 24 | Gemma 7B Open source Google DeepMind | 72.3% | — | 21 Feb 2024 | — | Source ↗ | |
| 25 | Qwen2.5-72B Open source Alibaba | 71.9% | — | 19 Sep 2024 | — | Source ↗ | |
| 26 | LLaMA-7B Open source Meta AI | 71.0% | — | 24 Feb 2023 | — | Source ↗ | |
| 27 | Llama 3-8B Open source Meta AI | 67.7% | — | 18 Apr 2024 | — | Source ↗ | |
| 28 | Falcon-7B Open source Technology Innovation Institute | 64.6% | — | 24 Apr 2023 | — | Source ↗ | |
| 29 | phi-3-mini 3.8B Open source Microsoft | 64.0% | — | 23 Apr 2024 | — | Source ↗ | |
| 30 | MPT-7B Open source MosaicML | 61.6% | — | 5 May 2023 | — | Source ↗ | |
| 31 | phi-3-small 7.4B Open source Microsoft | 58.1% | — | 23 Apr 2024 | — | Source ↗ | |
| 32 | Gemma 2B Open source Google DeepMind | 53.2% | — | 21 Feb 2024 | — | Source ↗ | |
| 33 | Phi-2 Open source Microsoft | 45.2% | — | 12 Dec 2023 | — | Source ↗ |
Caveats: settings (reasoning effort, agent scaffold) differ between rows and materially affect scores; the best reported setting per model is shown. Source: Epoch AI, CC BY 4.0.