Ironbee Raw Code Review - Full - F1: leaderboard
Ironbee raw code review (2026): standalone model review of real pull requests, three repeated runs. Full uses 50 PRs and Frontier 10 PRs; no cohort pooling. Model-specific requested tiers remain explicit; precision, recall, F1 and F2 are separate metrics. Aggregated source fractions are multiplied by 100, with full precision retained. F1 (%).
Metric: F1 (%). Source: raw.githubusercontent.com. Saturation forecast: Rough model projection: around 2026. 52 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Ironbee Study claude-sonnet-5.5 - low tier | 55.99 | |
| 2 | Ironbee Study grok-4.7 - low tier | 55.19 | |
| 3 | Ironbee Study claude-opus-5.5 - low tier | 55.09 | |
| 4 | Ironbee Study muse-spark-1.3 - low tier | 54.35 | |
| 5 | Ironbee Study gemini-4-argon - xhigh tier | 54.33 | |
| 6 | Ironbee Study muse-spark-1.3 - default tier | 52.3 | |
| 7 | Ironbee Study gemini-4-argon - default tier | 51.81 | |
| 8 | Ironbee Study gemini-4-argon - low tier | 51.69 | |
| 9 | Ironbee Study claude-opus-5.5 - default tier | 51.44 | |
| 10 | Ironbee Study muse-wonder - low tier | 50.74 |
Interactive version: theaggregate.ai/benchmark?slug=ironbee-raw-code-review-full-f1 · How It Works · Data refreshed daily, snapshot 2026-10-21.