M4Bench - Codex Native With-Skill: leaderboard
M4Bench tests whether coding agents can derive clinical data from operational database schemas. Results average 28 tasks and five trials per task, with and without supplied domain skills; Codex and Claude Code are measured separately. Task-balanced mean value-column match rate (0–1),28 tasks,5 trials,medium reasoning.
Metric: Task-balanced mean value-column match rate (0–1),28 tasks,5 trials,medium reasoning. Source: arxiv.org. Saturation forecast: Around January 2027. 2 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 (Medium) | 0.88 | #32 (GPT-5.5) |
| 2 | GPT-5.4 Mini (Medium) | 0.8 | #213 (GPT-5.4 Mini) |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=m4bench-codex-native-with-skill · How It Works · Data refreshed daily, snapshot 2026-10-29.