Scale AI - MCP Atlas: leaderboard

Evaluates real-world tool use through the Model Context Protocol. Part of Scale's SEAL agentic benchmark suite.

Metric: Pass Rate (self-reported). Source: benchmarklist.com. Status: saturation imminent. 51 models tracked.

Top models

#ModelScore
1Muse Spark 1.188.1
2Claude Opus 585.8
3Claude Opus 5 (xHigh)85.8
4Kimi K3 (Max)84.2
5Hy4 preview83.7
6Gemini 3.5 Flash83.6
7Gemini 3.5 Flash (High)83.6
8Claude Opus 4.8 (Max)83.6
9Claude Fable 583.3
10Claude Fable 5 (Max)83.3
11GPT-5.5 (xHigh)82.8
12Muse Spark82.2
13GPT-5.6 Sol81.8
14GPT-5.6 Sol (Max)81.8
15Seed 2.1 Turbo80.3

Interactive version: theaggregate.ai/benchmark?slug=scale-ai-mcp-atlas · How It Works · Data refreshed daily, snapshot 2026-09-05.