agents benchmark · included in ECI
LMCA
LMCA benchmark (score column: Score).
Results
55
Random baseline
0.0%
Score ceiling
85%
Released
12 Aug 2026
Score by model release date
Best score by organisation
Leaderboard
| # | Model | Score | Relative | Setting | Released | Run | Evidence |
|---|---|---|---|---|---|---|---|
| 1 | Kimi K3 Open source MoonshotAI | 52.7% | max | 16 Jul 2026 | — | Source ↗ | |
| 2 | DeepSeek V4.1 Flash (unknown) Open source DeepSeek | 47.0% | unknown | 9 Sep 2026 | — | Source ↗ | |
| 3 | GLM 5.2 Open source Z.ai | 45.8% | max | 16 Jun 2026 | — | Source ↗ | |
| 4 | DeepSeek V4 Pro 0813 (unknown) Open source DeepSeek | 45.5% | unknown | 13 Aug 2026 | — | Source ↗ | |
| 5 | DeepSeek V4 Flash 0731 Open source DeepSeek | 41.7% | max | 31 Jul 2026 | — | Source ↗ | |
| 6 | Qwen3.8 27B Open source Qwen | 41.4% | — | 14 Aug 2026 | — | Source ↗ | |
| 7 | DeepSeek v4 Pro (high) Open source DeepSeek | 41.2% | max | 24 Apr 2026 | — | Source ↗ | |
| 8 | Gemma 4 31B Open source Google | 39.3% | — | 2 Apr 2026 | — | Source ↗ | |
| 9 | Qwen3.5 397B A17B Open source Qwen | 37.9% | — | 13 Feb 2026 | — | Source ↗ | |
| 10 | Inkling (xhigh) Open source Thinking Machines | 37.6% | xhigh | 15 Jul 2026 | — | Source ↗ | |
| 11 | Kimi K2.6 Open source MoonshotAI | 37.3% | — | 20 Apr 2026 | — | Source ↗ | |
| 12 | Nemotron 3 Ultra Open source NVIDIA | 36.9% | — | 4 Jun 2026 | — | Source ↗ | |
| 13 | DeepSeek v4 Flash (max) Open source DeepSeek | 35.9% | max | 24 Apr 2026 | — | Source ↗ | |
| 14 | Qwen3.6 27B Open source Qwen | 34.5% | — | 22 Apr 2026 | — | Source ↗ | |
| 15 | Qwen3.5-27B Open source Qwen | 34.0% | — | 25 Feb 2026 | — | Source ↗ | |
| 16 | MiniMax M3 Open source MiniMax | 33.7% | — | 1 Jun 2026 | — | Source ↗ | |
| 17 | Qwen3.5-122B-A10B Open source Qwen | 32.2% | none | 25 Feb 2026 | — | Source ↗ | |
| 18 | Qwen3.6 35B A3B Open source Qwen | 29.7% | — | 27 Apr 2026 | — | Source ↗ | |
| 19 | Gemma 4 26B A4B Open source Google | 29.7% | — | 2 Apr 2026 | — | Source ↗ | |
| 20 | MiMo-V2.5-Pro Open source Xiaomi | 29.5% | — | 22 Apr 2026 | — | Source ↗ | |
| 21 | Qwen3.5-35B-A3B Open source Qwen | 29.5% | — | 24 Feb 2026 | — | Source ↗ | |
| 22 | Qwen3-235B-A22B-Thinking (Jul 2025) Open source Alibaba | 29.3% | — | 25 Jul 2025 | — | Source ↗ | |
| 23 | DeepSeek V3.2 Exp Open source DeepSeek | 29.1% | thinking | 29 Sep 2025 | — | Source ↗ | |
| 24 | DeepSeek-V3.2 (Thinking; Novita) Open source DeepSeek | 28.8% | — | 1 Dec 2025 | — | Source ↗ | |
| 25 | DeepSeek V3.1 Terminus Open source DeepSeek | 28.6% | — | 22 Sep 2025 | — | Source ↗ | |
| 26 | Mistral Medium 3.5 Open source Mistral | 26.1% | — | 28 Apr 2026 | — | Source ↗ | |
| 27 | Qwen3 235B A22B Open source Qwen | 25.0% | — | 28 Apr 2025 | — | Source ↗ | |
| 28 | Qwen3.5-9B Open source Qwen | 24.5% | — | 24 Feb 2026 | — | Source ↗ | |
| 29 | DeepSeek V3.1 Open source DeepSeek | 24.3% | thinking | 21 Aug 2025 | — | Source ↗ | |
| 30 | Qwen3-235B-A22B-Instruct (Jul 2025) Open source Alibaba | 23.5% | — | 25 Jul 2025 | — | Source ↗ | |
| 31 | Qwen3-30B-A3B-Instruct (Jul 2025) Open source Alibaba | 22.4% | — | 29 Jul 2025 | — | Source ↗ | |
| 32 | gpt-oss-120b Open source OpenAI | 22.1% | high | 5 Aug 2025 | — | Source ↗ | |
| 33 | Mistral Small 4 Open source Mistral | 20.6% | — | 16 Mar 2026 | — | Source ↗ | |
| 34 | Qwen3-30B-A3B-Thinking (Jul 2025) Open source Alibaba | 19.5% | — | 30 Jul 2025 | — | Source ↗ | |
| 35 | Qwen3 14B Open source Qwen | 18.2% | — | 29 Apr 2025 | — | Source ↗ | |
| 36 | Llama 3.3 70B Open source Meta AI | 17.5% | — | 6 Dec 2024 | — | Source ↗ | |
| 37 | Qwen3 32B Open source Qwen | 17.3% | — | 29 Apr 2025 | — | Source ↗ | |
| 38 | Mistral Large 3 Open source Mistral AI | 16.7% | — | 2 Dec 2025 | — | Source ↗ | |
| 39 | Llama 4 Maverick Open source Meta | 15.9% | — | 6 Apr 2025 | — | Source ↗ | |
| 40 | Qwen3 30B A3B Open source Qwen | 15.8% | — | 29 Apr 2025 | — | Source ↗ | |
| 41 | DeepSeek-V3 (Mar 2025) Open source DeepSeek | 15.5% | — | 24 Mar 2025 | — | Source ↗ | |
| 42 | DeepSeek V3.2 Open source DeepSeek | 15.2% | — | 1 Dec 2025 | — | Source ↗ | |
| 43 | Llama 3.1-70B Open source Meta AI | 14.8% | — | 23 Jul 2024 | — | Source ↗ | |
| 44 | gpt-oss-20b Open source OpenAI | 14.5% | high | 5 Aug 2025 | — | Source ↗ | |
| 45 | Qwen2.5-72B Open source Alibaba | 13.4% | — | 19 Sep 2024 | — | Source ↗ | |
| 46 | Gemma 3 27B Open source Google | 12.3% | — | 12 Mar 2025 | — | Source ↗ | |
| 47 | Llama 4 Scout Open source Meta | 12.0% | — | 5 Apr 2025 | — | Source ↗ | |
| 48 | Cohere Command A Open source Cohere | 10.3% | — | 13 Mar 2025 | — | Source ↗ | |
| 49 | Qwen3 8B Open source Qwen | 8.8% | — | 28 Apr 2025 | — | Source ↗ | |
| 50 | Gemma 2 27B Open source Google | 7.1% | — | 24 Jun 2024 | — | Source ↗ | |
| 51 | Qwen2.5-7B Open source Alibaba | 6.4% | — | 19 Sep 2024 | — | Source ↗ | |
| 52 | Llama 3.1-8B Open source Meta AI | 5.4% | — | 23 Jul 2024 | — | Source ↗ | |
| 53 | Command R+ Open source Cohere,Cohere Labs (formerly Cohere for AI) | 5.0% | — | 30 Aug 2024 | — | Source ↗ | |
| 54 | Gemma 3 12B Open source Google | 4.5% | — | 12 Mar 2025 | — | Source ↗ | |
| 55 | Gemma 3 4B Open source Google | 2.8% | — | 12 Mar 2025 | — | 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.