Toolathlon-Verified
Can the model plan, use tools, and finish a multi-step task?
Agents12 ranked models16 reported values3 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 | Claude Fable 5Anthropic | 77.9% | max effortMoonshot AI report1 value · 1 reportJul 17, 2026 · source |
| 2 | Claude Opus 4.8Anthropic | 76.2% | Reported configurationDeepSeek report3 values · 3 reportsJul 31, 2026 · source |
| 3 | Muse Spark 1.1Meta | 75.6% | xhigh effortMeta report1 value · 1 reportJul 9, 2026 · source |
| 4 | GPT-5.6 SolOpenAI | 74.9% | max effortMoonshot AI report1 value · 1 reportJul 17, 2026 · source |
| 5 | GPT-5.5OpenAI | 73.5% | xhigh effortMoonshot AI report2 values · 2 reportsJul 17, 2026 · source |
| 6 | Kimi K3Moonshot AI | 73.2% | max effortMoonshot AI report1 value · 1 reportJul 17, 2026 · source |
| 7 | DeepSeek V4 Flash 0731DeepSeek | 70.3% | max effortDeepSeek report1 value · 1 reportJul 31, 2026 · source |
| 8 | Gemini 3.1 ProGoogle | 61.1% | high effortMeta report1 value · 1 reportJul 9, 2026 · source |
| 9 | GLM-5.2Z.ai | 59.9% | Reported configurationDeepSeek report2 values · 2 reportsJul 31, 2026 · source |
| 10 | DeepSeek V4 ProDeepSeek | 55.9% | Reported configurationDeepSeek report1 value · 1 reportJul 31, 2026 · source |
| 11 | DeepSeek-V4-FlashDeepSeek | 49.7% | Reported configurationDeepSeek report1 value · 1 reportJul 31, 2026 · source |
| 12 | Muse Spark 1.0Meta | 49.4% | Reported configurationMeta report1 value · 1 reportJul 9, 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 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.
108-task current release using the official Toolathlon harness.
Method / source