Pokee-Isaac 28B Report - MCP-Atlas Coverage: leaderboard
Scale AI's MCP-Atlas, all 500 public tasks against a live 36-server MCP sandbox (GitHub, Slack, Google Workspace, Notion and others) with caches off, 40 tool turns and 4,096 output tokens. A shared GPT-5.5 judge scores the mean fraction of required claims satisfied (%). Pokee-Isaac ran at temperature 0.7, the other models at 1.0. From Pokee AI's Pokee-Isaac 28B technical report (September 10, 2026 revision): Pokee ran every listed model itself through hosted APIs under one harness, with each peer's default temperature (unless a benchmark section fixes one) and a high thinking level where available; exhausted retries, context overflows and verifier failures count as zero. Higher is better.
Metric: Mean claim coverage (%). Source: pokee.ai. Saturation forecast: Around February 2027. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.6 Luna | 77.9 | #117 |
| 2 | Gemini 3.5 Flash Lite | 76.67 | #300 |
| 3 | Qwen 3.5 122B A10B | 70.24 | #329 |
| 4 | Claude Haiku 4.5 | 56.45 | #568 |
| 5 | Nemotron 3 Super | 48.95 | #766 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=pokee-isaac-28b-report-mcp-atlas-coverage · How It Works · Data refreshed daily, snapshot 2026-11-12.