SWE-Review (MiniMax): leaderboard

Each model, run by MiniMax through Claude Code, reviews code containing a known defect; a review counts only if it finds that defect and reports nothing false, averaged over four runs. Unlike SWE-bench it grades code review rather than patch writing; the case count is not published.

Metric: Correct reviews (%). Source: github.com. Saturation forecast: Around January 2028. 5 models tracked. Updated 17 Nov 2026.

Top models

#ModelScoreOverall rank
1Claude Opus 4.516.2#161
2Claude Sonnet 4.510.5#258
3MiniMax-M2.18.9#573
4DeepSeek V3.26.4#417
5MiniMax-M23.4#687

No result here: #1 Claude Fable 5.2, #2 GPT-6.2 Astra, #3 Gemini 4 Carbon.

Interactive version: theaggregate.ai/benchmark?slug=swe-review-minimax · How It Works · Data refreshed daily, snapshot 2026-11-17.