reasoning benchmark · included in ECI
DTBench
DTBench benchmark (score column: Accuracy).
Results
71
Random baseline
40.0%
Score ceiling
100%
Released
12 Aug 2026
Score by model release date
Best score by organisation
Leaderboard
| # | Model | Score | Relative | Setting | Released | Run | Evidence |
|---|---|---|---|---|---|---|---|
| 1 | DeepSeek V4 Pro 0813 (unknown) Open source DeepSeek | 93.9% | unknown | 13 Aug 2026 | — | Source ↗ | |
| 2 | GLM 5.2 Open source Z.ai | 93.6% | max | 16 Jun 2026 | — | Source ↗ | |
| 3 | Kimi K3 Open source MoonshotAI | 91.2% | max | 16 Jul 2026 | — | Source ↗ | |
| 4 | DeepSeek V4 Flash 0731 Open source DeepSeek | 90.9% | max | 31 Jul 2026 | — | Source ↗ | |
| 5 | Kimi K2.6 Open source MoonshotAI | 90.9% | — | 20 Apr 2026 | — | Source ↗ | |
| 6 | DeepSeek v4 Pro (high) Open source DeepSeek | 90.7% | max | 24 Apr 2026 | — | Source ↗ | |
| 7 | Nemotron 3 Ultra Open source NVIDIA | 90.1% | — | 4 Jun 2026 | — | Source ↗ | |
| 8 | DeepSeek V4.1 Flash (unknown) Open source DeepSeek | 89.9% | unknown | 9 Sep 2026 | — | Source ↗ | |
| 9 | Qwen3.8 27B Open source Qwen | 88.0% | — | 14 Aug 2026 | — | Source ↗ | |
| 10 | DeepSeek V3.2 Exp Open source DeepSeek | 87.7% | thinking | 29 Sep 2025 | — | Source ↗ | |
| 11 | Inkling (xhigh) Open source Thinking Machines | 87.5% | xhigh | 15 Jul 2026 | — | Source ↗ | |
| 12 | Qwen3.5 397B A17B Open source Qwen | 87.5% | — | 13 Feb 2026 | — | Source ↗ | |
| 13 | DeepSeek v4 Flash (max) Open source DeepSeek | 86.4% | max | 24 Apr 2026 | — | Source ↗ | |
| 14 | DeepSeek-V3.2 (Thinking; Novita) Open source DeepSeek | 85.6% | — | 1 Dec 2025 | — | Source ↗ | |
| 15 | MiMo-V2.5-Pro Open source Xiaomi | 84.5% | — | 22 Apr 2026 | — | Source ↗ | |
| 16 | Qwen3.5-122B-A10B Open source Qwen | 84.3% | none | 25 Feb 2026 | — | Source ↗ | |
| 17 | Gemma 4 31B Open source Google | 82.7% | — | 2 Apr 2026 | — | Source ↗ | |
| 18 | DeepSeek V3.1 Open source DeepSeek | 82.7% | thinking | 21 Aug 2025 | — | Source ↗ | |
| 19 | Qwen3.5-27B Open source Qwen | 82.4% | — | 25 Feb 2026 | — | Source ↗ | |
| 20 | DeepSeek V3.1 Terminus Open source DeepSeek | 81.3% | — | 22 Sep 2025 | — | Source ↗ | |
| 21 | Qwen3-235B-A22B-Thinking (Jul 2025) Open source Alibaba | 80.3% | — | 25 Jul 2025 | — | Source ↗ | |
| 22 | Qwen3.5-35B-A3B Open source Qwen | 80.0% | — | 24 Feb 2026 | — | Source ↗ | |
| 23 | MiniMax M3 Open source MiniMax | 78.9% | — | 1 Jun 2026 | — | Source ↗ | |
| 24 | Qwen3-235B-A22B-Instruct (Jul 2025) Open source Alibaba | 78.4% | — | 25 Jul 2025 | — | Source ↗ | |
| 25 | Qwen3.6 27B Open source Qwen | 78.1% | — | 22 Apr 2026 | — | Source ↗ | |
| 26 | gpt-oss-120b Open source OpenAI | 76.3% | high | 5 Aug 2025 | — | Source ↗ | |
| 27 | Qwen3 235B A22B Open source Qwen | 75.7% | — | 28 Apr 2025 | — | Source ↗ | |
| 28 | Mistral Medium 3.5 Open source Mistral | 75.5% | — | 28 Apr 2026 | — | Source ↗ | |
| 29 | Gemma 4 26B A4B Open source Google | 74.9% | — | 2 Apr 2026 | — | Source ↗ | |
| 30 | Qwen3.6 35B A3B Open source Qwen | 73.9% | — | 27 Apr 2026 | — | Source ↗ | |
| 31 | Qwen3.5-9B Open source Qwen | 71.2% | — | 24 Feb 2026 | — | Source ↗ | |
| 32 | Mistral Small 4 Open source Mistral | 70.9% | — | 16 Mar 2026 | — | Source ↗ | |
| 33 | Qwen3-30B-A3B-Thinking (Jul 2025) Open source Alibaba | 69.3% | — | 30 Jul 2025 | — | Source ↗ | |
| 34 | gpt-oss-20b Open source OpenAI | 68.0% | high | 5 Aug 2025 | — | Source ↗ | |
| 35 | Qwen3 32B Open source Qwen | 67.5% | — | 29 Apr 2025 | — | Source ↗ | |
| 36 | Qwen3-30B-A3B-Instruct (Jul 2025) Open source Alibaba | 67.2% | — | 29 Jul 2025 | — | Source ↗ | |
| 37 | Mistral Large 3 Open source Mistral AI | 65.1% | — | 2 Dec 2025 | — | Source ↗ | |
| 38 | DeepSeek-V3 (Mar 2025) Open source DeepSeek | 64.8% | — | 24 Mar 2025 | — | Source ↗ | |
| 39 | Qwen3 14B Open source Qwen | 64.0% | — | 29 Apr 2025 | — | Source ↗ | |
| 40 | Qwen2.5-72B Open source Alibaba | 62.9% | — | 19 Sep 2024 | — | Source ↗ | |
| 41 | DeepSeek V3.2 Open source DeepSeek | 62.7% | — | 1 Dec 2025 | — | Source ↗ | |
| 42 | Llama 4 Maverick Open source Meta | 61.9% | — | 6 Apr 2025 | — | Source ↗ | |
| 43 | Llama 3.1-405B Open source Meta AI | 61.4% | — | 23 Jul 2024 | — | Source ↗ | |
| 44 | Cohere Command A Open source Cohere | 61.3% | — | 13 Mar 2025 | — | Source ↗ | |
| 45 | Magistral Small 1.2 Open source Mistral AI | 61.3% | — | 18 Sep 2025 | — | Source ↗ | |
| 46 | Mistral Large 2 (Jul 2024) Open source Mistral AI | 61.2% | — | 24 Jul 2024 | — | Source ↗ | |
| 47 | Mistral Large 2 (Nov 2024) Open source Mistral AI | 60.8% | — | 18 Nov 2024 | — | Source ↗ | |
| 48 | Qwen3 30B A3B Open source Qwen | 60.3% | — | 29 Apr 2025 | — | Source ↗ | |
| 49 | Llama 3.1-70B Open source Meta AI | 60.0% | — | 23 Jul 2024 | — | Source ↗ | |
| 50 | Mistral Small 3.2 Open source Mistral AI | 59.9% | — | 20 Jun 2025 | — | Source ↗ | |
| 51 | Qwen3 8B Open source Qwen | 59.7% | — | 28 Apr 2025 | — | Source ↗ | |
| 52 | Llama 3.3 70B Open source Meta AI | 59.5% | — | 6 Dec 2024 | — | Source ↗ | |
| 53 | Mistral Small 3.1 Open source Mistral AI | 58.6% | — | 17 Mar 2025 | — | Source ↗ | |
| 54 | Llama 4 Scout Open source Meta | 57.9% | — | 5 Apr 2025 | — | Source ↗ | |
| 55 | Mixtral 8x22B Open source Mistral AI | 55.1% | — | 17 Apr 2024 | — | Source ↗ | |
| 56 | Command R+ Open source Cohere,Cohere Labs (formerly Cohere for AI) | 54.9% | — | 30 Aug 2024 | — | Source ↗ | |
| 57 | Llama 3-70B Open source Meta AI | 54.2% | — | 18 Apr 2024 | — | Source ↗ | |
| 58 | Gemma 3 27B Open source Google | 52.5% | — | 12 Mar 2025 | — | Source ↗ | |
| 59 | Gemma 3 4B Open source Google | 50.9% | — | 12 Mar 2025 | — | Source ↗ | |
| 60 | Llama 3.1-8B Open source Meta AI | 50.9% | — | 23 Jul 2024 | — | Source ↗ | |
| 61 | Mixtral 8x7B Open source Mistral AI | 49.6% | — | 11 Dec 2023 | — | Source ↗ | |
| 62 | Mistral Small v24.09 Open source Mistral AI | 49.5% | — | 17 Sep 2024 | — | Source ↗ | |
| 63 | Gemma 3 12B Open source Google | 48.8% | — | 12 Mar 2025 | — | Source ↗ | |
| 64 | Mistral Nemo Open source Mistral | 48.6% | — | 18 Jul 2024 | — | Source ↗ | |
| 65 | Gemma 2 27B Open source Google | 48.0% | — | 24 Jun 2024 | — | Source ↗ | |
| 66 | Qwen2.5-7B Open source Alibaba | 47.7% | — | 19 Sep 2024 | — | Source ↗ | |
| 67 | Ministral 8B Open source Mistral AI | 45.7% | — | 16 Oct 2024 | — | Source ↗ | |
| 68 | Llama 3-8B Open source Meta AI | 43.9% | — | 18 Apr 2024 | — | Source ↗ | |
| 69 | Mistral 7B v0.3 Open source Mistral AI | 42.5% | — | 27 May 2024 | — | Source ↗ | |
| 70 | Llama 2-13B Open source Meta AI | 42.2% | — | 18 Jul 2023 | — | Source ↗ | |
| 71 | Llama 2-70B Open source Meta AI | 41.6% | — | 18 Jul 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.