Benchmark
Terminal-Bench
Agentic benchmarks measure how well models carry out tasks that take several steps. Scores are percentages. Higher is better.
- Models measured
- 61
- Scores recorded
- 67
- Sources
- 3
- IQ weight
- 1.73
Score distribution
- 41.8%
- Mean
- 39.5%
- Median
- 18.5%
- Std dev
- 21.8%
- 10th percentile
- 66.5%
- 90th percentile
- 3.4%
- Min
This benchmark has a weight of 1.73 for chat models, measured over 61 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.
Leaderboard
| # | Model | Score |
|---|---|---|
| 1 | GPT 5.5· Epoch AI Benchmarking Hub data | 84.7% |
| 2 | GPT 5.4· Epoch AI Benchmarking Hub data | 81.8% |
| 3 | Claude Opus 4.7· Epoch AI Benchmarking Hub data | 80.2% |
| 4 | Gemini 3.1 Pro Preview· Epoch AI Benchmarking Hub data | 80.2% |
| 5 | Claude Opus 4.6· Epoch AI Benchmarking Hub data | 79.8% |
| 6 | GPT-5.3 Codex· Epoch AI Benchmarking Hub data | 78.4% |
| 7 | Gemini 3 Pro Preview· Epoch AI Benchmarking Hub data | 69.4% |
| 8 | GPT-5.2 Codex· Epoch AI Benchmarking Hub data | 66.5% |
| 9 | GPT 5.2· Epoch AI Benchmarking Hub data | 64.9% |
| 10 | GPT 5.2medium· Epoch AI Benchmarking Hub data | 64.9% |
| 11 | Gemini 3 Flash· Epoch AI Benchmarking Hub data | 64.3% |
| 12 | Claude Opus 4.5· Epoch AI Benchmarking Hub data | 63.1% |
| 13 | GPT-5.1 Codex Mini· Epoch AI Benchmarking Hub data | 61.6% |
| 14 | GPT-5.1 Codex Max· Epoch AI Benchmarking Hub data | 60.4% |
| 15 | Claude Opus 4.5128k· Epoch AI Benchmarking Hub data | 59.1% |
| 16 | GPT-5.1 Codex· Epoch AI Benchmarking Hub data | 57.8% |
| 17 | Grok 4.20· Epoch AI Benchmarking Hub data | 57.3% |
| 18 | Claude Sonnet 4.6· Epoch AI Benchmarking Hub data | 53.4% |
| 19 | GLM 5· Epoch AI Benchmarking Hub data | 52.4% |
| 20 | Claude Sonnet 4.5· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claimValues from 2 sources
| 50% |
| 21 | GPT 5· Epoch AI Benchmarking Hub data | 49.6% |
| 22 | GPT 5medium· Epoch AI Benchmarking Hub data | 49.6% |
| 23 | MiniMax M2.1· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claimValues from 2 sources
| 47.9% |
| 24 | GPT 5.1medium· Epoch AI Benchmarking Hub data | 47.6% |
| 25 | GPT 5.1· Epoch AI Benchmarking Hub data | 47.6% |
| 26 | Kimi K2 Thinking· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claimValues from 2 sources
| 47.1% |
| 27 | MiniMax M2· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it.Values from 2 sources
| 46.3% |
| 28 | MiniMax M2.7· Epoch AI Benchmarking Hub data | 45.1% |
| 29 | GPT-5 Codex· Epoch AI Benchmarking Hub data | 44.3% |
| 30 | Claude Opus 4.1· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claimValues from 2 sources
| 43.3% |
| 31 | Kimi K2.5· Epoch AI Benchmarking Hub data | 43.2% |
| 32 | MiniMax M2.5· Epoch AI Benchmarking Hub data | 42.7% |
| 33 | GLM 5.3· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim | 41.2% |
| 34 | Claude Haiku 4.5· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claimValues from 2 sources
| 41% |
| 35 | GLM 4.6· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claimValues from 2 sources
| 40.5% |
| 36 | DeepSeek V3.2· Epoch AI Benchmarking Hub data | 39.6% |
| 37 | LongCat Flash Chat· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 39.5% |
| 38 | Claude Opus 4· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 39.2% |
| 39 | DeepSeek V3.2 Exp· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 37.7% |
| 40 | GLM 4.5· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 37.5% |
| 41 | Claude Sonnet 4· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 35.5% |
| 42 | GLM 5.3 Flash· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim | 35.5% |
| 43 | Claude Sonnet 3.7· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 35.2% |
| 44 | GPT-5 mini· Epoch AI Benchmarking Hub data | 34.8% |
| 45 | longcat-flash-lite· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 33.8% |
| 46 | GLM 4.7· Epoch AI Benchmarking Hub dataValues from 2 sources
| 33.4% |
| 47 | Gemini 2.5 Pro· Epoch AI Benchmarking Hub data | 32.6% |
| 48 | Nova 2 Lite· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 32.5% |
| 49 | GPT-5 minimedium· Epoch AI Benchmarking Hub data | 31.9% |
| 50 | DeepSeek V3.1· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 31.3% |
Showing 1–50 of 67 results.
Data credits: Data by LLM Stats · Epoch AI Benchmarking Hub, CC BY 4.0 (licence) · Hugging Face official benchmark leaderboards; licence varies by benchmark