Benchmarks · Agents & computer use · Coding
Terminal-Bench 4.1
66 professional computer-work tasks done by an agent in a terminal, with up to 8 hours and 5 trials per task. Share solved; rows are harness-plus-model pairs. 4.1 re-grades 4.0's runs with checkers that resist cheating and wrongly fail fewer runs.
37 models · 51 results · data as of 2026-11-12 · tbench.ai ↗
Leaderboard
Fable 5.2 max
Heavy line: best score at each price. Thin lines join one model's effort levels. Cost uses each model's median subscription plan, or API price with no plan data. 22 more models under All models in settings.
More from this benchmark
2 published measuresTime per task51 results · lower is better
21 more models under All models in settings.
Output tokens per task51 results · lower is better
All results
51 rows| # | Model | Effort | Terminal-Bench 4.1 | Subscription $ / task | API $ / task | Time | Output tokens | Harness |
|---|---|---|---|---|---|---|---|---|
| 1 | Fable 5.2 | max | 65.8% ±2.8 | $1.84 | $50 | 2.0 h | 491k | Claude Code |
| 2 | Opus 5.5 | max | 65.8% ±3.1 | $0.22 | $14 | 1.3 h | 269k | Claude Code |
| 3 | Sonnet 5.5 | max | 63.9% ±3.0 | $0.34 | $22 | 2.1 h | 565k | Claude Code |
| 4 | GPT-6.1 Sol | max | 60.9% ±3.0 | $0.064 | $1.92 | 53 min | 95k | Codex |
| 5 | GPT-6 Astra | max | 60.3% ±2.8 | $0.42 | $9.90 | 47 min | 73k | Codex |
| 6 | GPT-6 Astra | high | 60.0% ±2.8 | $0.29 | $6.88 | 35 min | 42k | Codex |
| 7 | GPT-6 Astra | xhigh | 59.7% ±2.7 | $0.30 | $7.12 | 37 min | 45k | Codex |
| 8 | Kimi K3.1 | max | 59.7% ±4.2 | $3.29 | $3.29 | 2.1 min | 81k | Claude Code |
| 9 | Fable 5.1 | xhigh | 57.6% ±3.3 | $2.30 | $15 | 53 min | 153k | Claude Code |
| 10 | Fable 5.1 | max | 57.3% ±3.8 | $2.95 | $19 | 1.1 h | 191k | Claude Code |
| 11 | GPT-6 Astra | medium | 55.1% ±2.7 | $0.25 | $5.80 | 31 min | 33k | Codex |
| 12 | Fable 5.1 | high | 54.9% ±3.4 | $1.88 | $12 | 50 min | 113k | Claude Code |
| 13 | Fable 5.1 | medium | 53.9% ±3.4 | $1.34 | $8.58 | 42 min | 79k | Claude Code |
| 14 | Opus 5 | xhigh | 53.3% ±3.1 | $0.32 | $18 | 1.3 h | 179k | Claude Code |
| 15 | GPT-6.1 Terra | max | 53.3% ±3.1 | $0.10 | $2.65 | 2.1 min | 100k | Codex |
| 16 | Opus 5 | max | 52.1% ±3.4 | $0.31 | $18 | 1.3 h | 200k | Claude Code |
| 17 | GPT-6 Astra | low | 50.9% ±2.8 | $0.20 | $4.72 | 28 min | 24k | Codex |
| 18 | GPT-6 Sol | max | 50.3% ±3.2 | $0.14 | $4.40 | 44 min | 93k | Codex |
| 19 | Opus 5 | high | 49.7% ±3.7 | $0.24 | $14 | 1.1 h | 144k | Claude Code |
| 20 | Opus 5 | medium | 45.1% ±3.9 | $0.17 | $9.67 | 55 min | 101k | Claude Code |
| 21 | Fable 5 | max | 44.9% ±3.8 | $2.48 | $22 | 1.2 h | 178k | Claude Code |
| 22 | Fable 5.1 | low | 43.3% ±3.6 | $1.11 | $7.15 | 38 min | 60k | Claude Code |
| 23 | GLM 5.3 | max | 41.2% ±3.3 | $0.86 | $8.27 | 1.6 h | 208k | Claude Code |
| 24 | Haiku 5.5 | max | 40.0% ±3.6 | $0.061 | $2.89 | 2.3 min | 386k | Claude Code |
| 25 | Grok 4.7 | xhigh | 38.8% ±3.6 | $1.04 | $11 | 1.6 h | 70k | Grok Build |
| 26 | DeepSeek V4.1 Flash 1017 | max | 38.5% ±3.8 | free | $0.56 | 2.5 min | 497k | Claude Code |
| 27 | GPT-5.6 Sol | max | 38.5% ±3.8 | $0.21 | $7.70 | 40 min | 71k | Codex |
| 28 | Hy4 | max | 37.9% ±3.5 | $0.30 | $1.05 | 2.1 min | 369k | Claude Code |
| 29 | GLM 5.3 Flash | max | 37.0% ±3.6 | free | $1.14 | 2.0 min | 460k | Claude Code |
| 30 | GLM 5.3 Flash | none | 35.5% ±3.6 | free | $0.36 | 1.7 h | 135k | Claude Code |