Benchmarks / Agentic
MCP Atlas (Scale AI)
How reliably a model completes realistic multi-step workflows over real Model Context Protocol servers: discovering the right tool in a noisy tool menu, calling it with correct parameters, recovering from errors and synthesising the results into an accurate final answer. Maintained by Scale AI. Score: Pass rate in percent over all 1,000 tasks; the published ± is the confidence half-width.
MCP Atlas (Scale AI)
Higher is better · best effort per model
full table below
| # | Model | Effort | Score |
|---|---|---|---|
| 1 | Muse Spark 1.1 | max | 88.1% |
| 2 | Fable 5.1 | max | 87.2% |
| 3 | Opus 5 | xhigh | 85.8% |
| 4 | Qwen 3.8 | xhigh | 84.5% |
| 5 | GLM 5.3 | max | 84.2% |
| 6 | Gemini 3.5 Flash | high | 83.6% |
| 7 | Fable 5 | max | 83.3% |
| 8 | Kimi K3 | max | 82.3% |
| 9 | Muse Spark | – | 82.2% |
| 10 | Opus 4.8 | max | 82.2% |
| 11 | GPT-5.6 Sol | max | 81.8% |
| 12 | Inkling Small | xhigh | 79.2% |
| 13 | Opus 4.7 | max | 79.1% |
| 14 | Gemini 3.1 Pro | high | 78.2% |
| 15 | Opus 4.6 | max | 76.8% |
| 16 | Inkling | xhigh | 76.0% |
| 17 | GPT-5.5 | xhigh | 75.3% |
| 18 | GPT-5.4 | xhigh | 70.6% |
| 19 | Gemini 3 Pro | high | 70.3% |
| 20 | Opus 4.5 | high | 69.8% |
| 21 | Sonnet 4.6 | max | 69.5% |
| 22 | GPT-5.2 | xhigh | 67.6% |
| 23 | Nemotron 3 Ultra | max | 63.1% |
| 24 | Gemini 3 Flash | high | 62.0% |
| 25 | Sonnet 4.5 | high | 59.5% |
| 26 | Gemini 3.1 Flash-Lite | high | 57.1% |
| 27 | GPT-5.4 Mini | xhigh | 56.7% |
| 28 | GPT-5.1 | high | 50.1% |
| 29 | o3 pro | high | 44.5% |
| 30 | Haiku 4.5 | xhigh | 40.2% |