AutomationBench
Can the model plan, use tools, and finish a multi-step task?
Top rankings
One row per model, using its best reported score across effort settings.
| Rank | Model | Best score | Best reported setting |
|---|---|---|---|
| 1 | Claude Fable 5.1Anthropic | 31.4% | Reported configurationAnthropic report1 value · 1 reportSep 1, 2026 · source |
| 2 | Claude Opus 5Anthropic | 26.9% | Reported configurationAnthropic report1 value · 1 reportSep 1, 2026 · source |
| 3 | GPT-5.6 SolOpenAI | 19.6% | Reported configurationAnthropic report1 value · 1 reportSep 1, 2026 · source |
| 4 | Claude Fable 5Anthropic | 17.1% | Reported configurationAnthropic report1 value · 1 reportSep 1, 2026 · source |
Effort curve
Every sourced cost-linked effort value for this exact version. Lines connect complete sweeps only.
Definition and comparison boundarypublic methodology
Can the model plan, use tools, and finish a multi-step task?
Agentic execution across tools, environments, or long-running workflows. Higher is better. The value is the percentage reported in this lab's table.
Harness, tools, prompts, attempt count, and benchmark version can materially change the result. Compare within one reporting context.
The launch table does not print a task-set revision or cost-per-task series. Anthropic states that production safeguards were enabled for these comparisons; safeguard interventions can lower scores and intervention cases can be scored as zero. The launch table does not print a task-set revision or cost-per-task series.
Method / source