Agentic / terminal scores
End-to-end tasks an agent completes in a terminal. Scores are per model and harness.
23 results on 1 board · checked
How this category is sourced
End-to-end tasks an agent completes in a terminal. Scores are per model and harness.
- Every value is copied as the board publishes it, with its metric, harness, as-of date and a source link. Boards use different metrics, so numbers are never compared across boards.
- No Ratelimit Now score: we do not normalise, weight or combine results.
- Stale means the as-of date is more than 90 days old. Frozen means the board stopped scoring new models. Preliminary is the board's own label.
- A model's score depends on the harness or agent it ran in. It is not a score for any tool on this site.
- Open and Chinese models sit in the same tables, with the license the board lists. Hosting region is not confirmed for any model.
Latest generation on this page
GPT-6.1 SolTerminal-Bench 4.0GPT-6 LunaTerminal-Bench 4.0GPT-6 AstraTerminal-Bench 4.0Qwen 3.8 Max 0902Terminal-Bench 4.0Muse Spark 1.3Terminal-Bench 4.0Grok 4.7Terminal-Bench 4.0GLM-5.3 FlashTerminal-Bench 4.0GLM-5.3Terminal-Bench 4.0Gemini 3.8 FlashTerminal-Bench 4.0Claude Sonnet 5.5Terminal-Bench 4.0Claude Opus 5.5Terminal-Bench 4.0Claude Fable 5.1Terminal-Bench 4.0
Not on these boards yet: GPT-6 Sol (see their model pages for other categories).
Terminal-Bench 4.0
Resolution rate (%)Harness: varies by row (5)Newest as-of 2026-09-2912 latest-gen · 12 current · 11 older · 2 open weights
Chart data as a table
| Model | Lab | Resolution rate (%) |
|---|---|---|
| Opus 5.5 (max) | Anthropic | 64.85 |
| Sonnet 5.5 (max) | Anthropic | 61.82 |
| GPT-6 Astra (max) | OpenAI | 58.18 |
| GPT-6.1 Sol (max) | OpenAI | 58.18 |
| Fable 5.1 (max) | Anthropic | 57.88 |
| GLM-5.3 (max) | Chinese labs | 41.82 |
| Grok 4.7 (xhigh) | xAI | 37.58 |
| GLM-5.3-Flash (none) | Chinese labs | 35.76 |
| Qwen3.8-Max-0902 (max) | Chinese labs | 26.97 |
| Gemini 3.8 Flash (high) | 19.09 |
Source: tbench.ai
| Lab | Rows |
|---|---|
| OpenAI | 2 |
| Anthropic | 3 |
| 1 | |
| xAI | 1 |
| Chinese labs | 3 |
Source: tbench.ai
| Model (as listed) | Resolution rate | Harness | License | As of | Source |
|---|---|---|---|---|---|
| Opus 5.5 (max)Anthropic · latest gen | 64.85% | Claude Code | Not listed | ↗ | |
| Sonnet 5.5 (max)Anthropic · latest gen | 61.82% | Claude Code | Not listed | ↗ | |
| GPT-6 Astra (max)OpenAI · latest gen | 58.18% | Codex | Not listed | ↗ | |
| GPT-6.1 Sol (max)OpenAI · latest gen | 58.18% | Codex | Not listed | ↗ | |
| Fable 5.1 (max)Anthropic · latest gen | 57.88% | Claude Code | Not listed | ↗ | |
| Opus 5 (xhigh)Anthropic · older version | 53.94% | Claude Code | Not listed | Stale | ↗ |
| GPT-6 Sol (max)OpenAI · older version | 49.39% | Codex | Not listed | ↗ | |
| Fable 5 (max)Anthropic · older version | 44.55% | Claude Code | Not listed | Stale | ↗ |
| GLM-5.3 (max)Z.ai · latest gen | 41.82% | Claude Code | Open weightsglm-5.3 (custom; LMArena l… | ↗ | |
| Grok 4.7 (xhigh)SpaceXAI · latest gen | 37.58% | Grok Build | Not listed | ↗ | |
| GPT-5.6 Sol (max)OpenAI · older version | 37.27% | Codex | Not listed | Stale | ↗ |
| GLM-5.3-Flash (none)Z.ai · latest gen | 35.76% | Claude Code | Open weightsmit | ↗ | |
| Qwen3.8-Max-0902 (max)Alibaba · latest gen | 26.97% | Claude Code | Not listed | ↗ | |
| Opus 4.8 (max)Anthropic · older version | 23.64% | Claude Code | Not listed | Stale | ↗ |
| GPT-5.6 Terra (max)OpenAI · older version | 21.52% | Codex | Not listed | Stale | ↗ |
| Grok 4.6 (high)SpaceXAI · older version | 20.3% | Grok Build | Not listed | ↗ | |
| Gemini 3.8 Flash (high)Google · latest gen | 19.09% | mini-SWE-agent | Not listed | ↗ | |
| GPT-5.6 Luna (max)OpenAI · older version | 17.27% | Codex | Not listed | Stale | ↗ |
| GPT-6 Luna (max)OpenAI · latest gen | 16.36% | Codex | Not listed | ↗ | |
| Muse Spark 1.3 (xhigh)Meta · latest gen | 14.55% | Muse Code | Not listed | ↗ | |
| Grok 4.5 (high)SpaceXAI · older version | 12.42% | Grok Build | Not listed | Stale | ↗ |
| Sonnet 5 (max)Anthropic · older version | 12.42% | Claude Code | Not listed | Stale | ↗ |
| Gemini 3.7 Flash (high)Google · older version | 11.21% | mini-SWE-agent | Not listed | ↗ |