AI model rankings
Understand capability differences before choosing AI resources.
Terminal-Bench 4.0 · Agent practice
Public evaluation snapshots · Ranked separately by dimension
Terminal-Bench 4.0 · Agent 任务完成率 · 版本 Terminal-Bench 4.0 · Scores range from 0–100; higher is better. Missing scores are excluded, and ties share a rank. Search preserves original ranks.
| Rank | Model and agent configuration | Score | Evidence type |
|---|---|---|---|
| 1 | anthropic/claude-opus-5-5 · [email protected] · maxAgent 实战 · claude-code 2.1.284 · 推理 max · Model release date 2026-09-22Model details ↗ | 64.8 | Leaderboard listing |
| 2 | anthropic/claude-sonnet-5-5 · [email protected] · maxAgent 实战 · claude-code 2.1.284 · 推理 max · Model release date 2026-09-28Model details ↗ | 61.8 | Leaderboard listing |
| 3 | openai/gpt-6.1-sol · [email protected] · maxAgent 实战 · codex 0.159.0 · 推理 max · Model release date 2026-09-29Model details ↗ | 58.2 | Leaderboard listing |
| 4 | anthropic/claude-fable-5-1 · [email protected] · maxAgent 实战 · claude-code 2.1.257 · 推理 max · Model release date 2026-09-01Model details ↗ | 57.9 | Leaderboard listing |
| 5 | anthropic/claude-opus-5 · [email protected] · maxAgent 实战 · claude-code 2.1.231 · 推理 max · Model release date 2026-07-24Model details ↗ | 51.8 | Leaderboard listing |
| 6 | openai/gpt-6-sol · [email protected] · maxAgent 实战 · codex 0.159.0 · 推理 max · Model release date 2026-09-22Model details ↗ | 49.4 | Leaderboard listing |
| 7 | anthropic/claude-fable-5 · [email protected] · maxAgent 实战 · claude-code 2.1.231 · 推理 max · Model release date 2026-06-09Model details ↗ | 44.6 | Leaderboard listing |
| 8 | anthropic/glm-5.3 · [email protected] · maxAgent 实战 · claude-code 2.1.207 · 推理 max · Model release date 2026-08-14Model details ↗ | 41.8 | Leaderboard listing |
| 9 | openai/gpt-5.6-sol · [email protected] · maxAgent 实战 · codex 0.149.1 · 推理 max · Model release date 2026-06-26Model details ↗ | 37.3 | Leaderboard listing |
| 10 | z-ai/glm-5.3-flash · [email protected] · noneAgent 实战 · claude-code 2.1.285 · 推理 none · Model release date 2026-08-26Model details ↗ | 35.8 | Leaderboard listing |
| 11 | qwen/qwen3.8-max-0902 · [email protected] · maxAgent 实战 · claude-code 2.1.285 · 推理 max · Model release date 2026-09-02Model details ↗ | 27.0 | Leaderboard listing |
| 12 | anthropic/claude-opus-4-8 · [email protected] · maxAgent 实战 · claude-code 2.1.231 · 推理 max · Model release date 2026-05-28Model details ↗ | 23.6 | Leaderboard listing |
| 13 | openai/gpt-5.6-terra · [email protected] · maxAgent 实战 · codex 0.149.1 · 推理 max · Model release date 2026-06-26Model details ↗ | 21.5 | Leaderboard listing |
| 14 | xai/grok-4.6 · [email protected] · noneAgent 实战 · grok-build 1.0.5 · 推理 none · Model release date 2026-08-12Model details ↗ | 20.3 | Leaderboard listing |
| 15 | gemini/gemini-3.8-flash · [email protected] · highAgent 实战 · mini-swe-agent 2.4.6 · 推理 high · Model release date 2026-09-02Model details ↗ | 19.1 | Leaderboard listing |
| 16 | openai/gpt-5.6-luna · [email protected] · maxAgent 实战 · codex 0.149.1 · 推理 max · Model release date 2026-06-26Model details ↗ | 17.3 | Leaderboard listing |
| 17 | openai/gpt-6-luna · [email protected] · maxAgent 实战 · codex 0.159.0 · 推理 max · Model release date 2026-09-22Model details ↗ | 16.4 | Leaderboard listing |
| 18 | meta/muse-spark-1.3 · [email protected] · xhighAgent 实战 · muse-code 1.4.1 · 推理 xhigh · Model release date 2026-09-02Model details ↗ | 14.6 | Leaderboard listing |
| 19 | anthropic/claude-sonnet-5 · [email protected] · maxAgent 实战 · claude-code 2.1.231 · 推理 max · Model release date 2026-06-30Model details ↗ | 12.4 | Leaderboard listing |
| 19 | xai/grok-4.5 · [email protected] · noneAgent 实战 · grok-build 1.0.5 · 推理 none · Model release date 2026-07-16Model details ↗ | 12.4 | Leaderboard listing |
| 21 | gemini/gemini-3.7-flash · [email protected] · highAgent 实战 · mini-swe-agent 2.4.6 · 推理 high · Model release date 2026-08-13Model details ↗ | 11.2 | Leaderboard listing |
Source evidence and evaluation methodology
Source:Terminal-Bench 4.0 · Agent practice · Apache-2.0. We organize original scores without creating cross-benchmark composite scores. Listing a result does not mean we independently reproduced it.
Attribution: Harbor Framework / Terminal-Bench contributors. Original submissions use Apache License 2.0. We show task completion for model, agent version, and inference configurations, not overall model-only capability. License and attribution for this source version
Source snapshot:1000b0941a244800a8ff48d461934c00 · Source commit：e67a2cb7db4d9795c212909b3395d8c22ad60aec · Original content SHA-256:29b754b80fdd1bcea650cff92d9d3401a9114fc968affa222c2a91985ef90c2f
Fetched on demand and published after admin review, not in real time. Undisclosed evaluation versions are not inferred.Different dimensions and test settings cannot be compared directly.