# Terminal-Bench 4.0

> The current Terminal-Bench release measures difficult computer work after recalibrating task resources, fixing unstable tasks, and removing tasks that no longer separate frontier systems.

Canonical page: https://benchlm.ai/benchmarks/terminal-bench-4

- Category: [Agentic](/agentic)
- Last updated: October 15, 2026 snapshot

## About Terminal-Bench 4.0

- Year: 2026
- Tasks: 66 professional computer-work tasks
- Format: Task completion rate across 5 trials per task
- Difficulty: Frontier autonomous knowledge work
- Paper: [Terminal-Bench 4.0](https://www.tbench.ai/news/terminal-bench-4-0)

Version 4.0 contains 66 tasks, gives every agent up to eight hours, fixes 19 tasks, and removes eight tasks for saturation, refusals, public solutions, or unresolved quality and platform issues. Those breaking changes required fresh trials, so 4.0 scores are not directly comparable with 3.0 or 2.x. Each row still combines a model with an agent harness.

The mirrored table is for reference; the scores BenchLM stores for each model feed the ranking. BenchAlign v5.8 gives Terminal-Bench 4.0 8% of the Agentic reference weight, so it moves the Agentic leaderboard and the overall ranking. Reference weights are relative weights in the calibrated model, not fixed shares of a score.

## Leaderboard (35 models)

| Rank | Model | Configuration | Creator | Score |
|------|-------|---------------|---------|-------|
| 1 | [Claude Opus 5.5](/models/claude-opus-5-5) | Claude Code · max reasoning | Anthropic | 64.85% |
| 2 | [Claude Sonnet 5.5](/models/claude-sonnet-5-5) | Claude Code · max reasoning | Anthropic | 61.82% |
| 3 | [GPT-6 Astra](/models/gpt-6-astra) | Codex · max reasoning | OpenAI | 58.18% |
| 4 | [GPT-6.1 Sol](/models/gpt-6-1-sol) | Codex · max reasoning | OpenAI | 58.18% |
| 5 | [Claude Fable 5.1](/models/claude-fable-5-1) | Claude Code · max reasoning | Anthropic | 57.88% |
| 6 | [GPT-6 Astra](/models/gpt-6-astra) | Codex · xhigh reasoning | OpenAI | 57.88% |
| 7 | [GPT-6 Astra](/models/gpt-6-astra) | Codex · high reasoning | OpenAI | 57.88% |
| 8 | [Claude Fable 5.1](/models/claude-fable-5-1) | Claude Code · xhigh reasoning | Anthropic | 57.88% |
| 9 | [Claude Fable 5.1](/models/claude-fable-5-1) | Claude Code · high reasoning | Anthropic | 54.55% |
| 10 | [GPT-6 Astra](/models/gpt-6-astra) | Codex · medium reasoning | OpenAI | 54.24% |
| 11 | [Claude Fable 5.1](/models/claude-fable-5-1) | Claude Code · medium reasoning | Anthropic | 53.94% |
| 12 | [Claude Opus 5](/models/claude-opus-5) | Claude Code · xhigh reasoning | Anthropic | 53.94% |
| 13 | [Claude Opus 5](/models/claude-opus-5) | Claude Code · max reasoning | Anthropic | 51.82% |
| 14 | [GPT-6 Astra](/models/gpt-6-astra) | Codex · low reasoning | OpenAI | 50.61% |
| 15 | [Claude Opus 5](/models/claude-opus-5) | Claude Code · high reasoning | Anthropic | 50.30% |
| 16 | [GPT-6 Sol](/models/gpt-6-sol) | Codex · max reasoning | OpenAI | 49.39% |
| 17 | [Claude Opus 5](/models/claude-opus-5) | Claude Code · medium reasoning | Anthropic | 44.85% |
| 18 | [Claude Fable 5](/models/claude-fable) | Claude Code · max reasoning | Anthropic | 44.55% |
| 19 | [Claude Fable 5.1](/models/claude-fable-5-1) | Claude Code · low reasoning | Anthropic | 43.33% |
| 20 | [GLM-5.3](/models/glm-5-3) | Claude Code · max reasoning | Z.AI | 41.82% |
| 21 | [Grok 4.7](/models/grok-4-7) | Grok Build · xhigh reasoning | xAI | 37.58% |
| 22 | [GPT-5.6 Sol](/models/gpt-5-6-sol) | Codex · max reasoning | OpenAI | 37.27% |
| 23 | [GLM-5.3-Flash](/models/glm-5-3-flash) | Claude Code · none reasoning | Z.AI | 35.76% |
| 24 | [Claude Opus 5](/models/claude-opus-5) | Claude Code · low reasoning | Anthropic | 34.85% |
| 25 | [Qwen3.8-Max-0902](https://hub.harborframework.com/datasets/terminal-bench/terminal-bench/latest?tab=leaderboard&leaderboard=4-0-0) | Claude Code · max reasoning | Alibaba | 26.97% |
| 26 | [Claude Opus 4.8](/models/claude-opus-4-8) | Claude Code · max reasoning | Anthropic | 23.64% |
| 27 | [GPT-5.6 Terra](/models/gpt-5-6-terra) | Codex · max reasoning | OpenAI | 21.52% |
| 28 | [Grok 4.6](/models/grok-4-6) | Grok Build · high reasoning | xAI | 20.30% |
| 29 | [Gemini 3.8 Flash](/models/gemini-3-8-flash) | mini-SWE-agent · high reasoning | Google | 19.09% |
| 30 | [GPT-5.6 Luna](/models/gpt-5-6-luna) | Codex · max reasoning | OpenAI | 17.27% |
| 31 | [GPT-6 Luna](/models/gpt-6-luna) | Codex · max reasoning | OpenAI | 16.36% |
| 32 | [Muse Spark 1.3](/models/muse-spark-1-3) | Muse Code · xhigh reasoning | Meta | 14.55% |
| 33 | [Grok 4.5](/models/grok-4-5) | Grok Build · high reasoning | xAI | 12.42% |
| 34 | [Claude Sonnet 5](/models/claude-sonnet-5) | Claude Code · max reasoning | Anthropic | 12.42% |
| 35 | [Gemini 3.7 Flash](/models/gemini-3-7-flash) | mini-SWE-agent · high reasoning | Google | 11.21% |

## FAQ

### What does Terminal-Bench 4.0 measure?

The current Terminal-Bench release measures difficult computer work after recalibrating task resources, fixing unstable tasks, and removing tasks that no longer separate frontier systems.

### Which model leads the published Terminal-Bench 4.0 snapshot?

Claude Opus 5.5 currently leads the published Terminal-Bench 4.0 snapshot with a score of 64.85%.

### How many models are evaluated on Terminal-Bench 4.0?

The October 15, 2026 snapshot contains 35 AI models.

### Does Terminal-Bench 4.0 affect BenchLM's overall score?

Yes. BenchAlign v5.8 gives Terminal-Bench 4.0 8% of the Agentic reference weight, so it moves the Agentic leaderboard and the overall ranking. Reference weights are relative weights in the calibrated model, not fixed shares of a score.