AgentWorldBench
Can the model plan, use tools, and finish a multi-step task?
Agents4 ranked models5 reported values1 reportsHigher is better
Top rankings
One row per model, using its best reported score across effort settings.
| Rank | Model | Best score | Best reported setting |
|---|---|---|---|
| 1 | Qwen-AgentWorld-397B-A17BQwen | 58.71 | Reported configurationQwen report1 value · 1 reportJun 22, 2026 · source |
| 2 | GPT-5.4OpenAI | 58.25 | Reported configurationQwen report1 value · 1 reportJun 22, 2026 · source |
| 3 | Qwen3.5-35B-A3BQwen | 56.39 | Reported configurationQwen report2 values · 1 reportJun 22, 2026 · source |
| 4 | Claude Sonnet 4.6Anthropic | 56.04 | Reported configurationQwen report1 value · 1 reportJun 22, 2026 · source |
Effort curve
Every sourced cost-linked effort value for this exact version. Lines connect complete sweeps only.
No cost-linked effort sweep for this version.
Definition and comparison boundarylimited 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. This benchmark reports points rather than percent correct.
Harness, tools, prompts, attempt count, and benchmark version can materially change the result. Compare within one reporting context.
Five-dimensional rubric mean across Qwen's seven-domain AgentWorldBench; Qwen states the 35B pipeline result as 47.73 → 56.39.
Method / source