SEAL - MCP Atlas: leaderboard
1,000 human-authored tool-use tasks spanning 36 real MCP servers and 220 tools, evaluating multi-step workflow execution via the Model Context Protocol.
Metric: Score. Source: scale.com. Source: benchmarklist.com. Saturation forecast: Around January 2027. 34 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Muse Spark 1.1 | 88.1 | #52 |
| 2 | Claude Fable 5.1 | 87.2 | #9 |
| 3 | Claude Opus 5 (Max) | 87 | #12 (Claude Opus 5) |
| 4 | GLM-5.3 (Max) | 86.8 | #39 (GLM-5.3) |
| 5 | Claude Opus 5 (xHigh) | 85.8 | #12 (Claude Opus 5) |
| 6 | Qwen 3.8 2.4T A95B (xHigh) | 84.5 | |
| 7 | GLM-5.3 | 84.2 | #39 |
| 8 | Hy4 preview | 83.7 | #38 |
| 9 | Gemini 3.5 Flash (High) | 83.6 | #60 (Gemini 3.5 Flash) |
| 10 | Claude Fable 5 (Max) | 83.3 | #16 (Claude Fable 5) |
| 11 | Kimi K3 (Max) | 82.3 | #30 (Kimi K3) |
| 12 | Claude Opus 4.8 (Max) | 82.2 | #37 (Claude Opus 4.8) |
| 13 | Muse Spark | 82.2 | #64 |
| 14 | GPT-5.6 Sol (Max) | 81.8 | #19 (GPT-5.6 Sol) |
| 15 | Seed 2.1 Turbo | 80.3 | #86 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=seal-mcp-atlas · How It Works · Data refreshed daily, snapshot 2026-10-20.