Active-SWE - Revealed: leaderboard
Active-SWE revealed potential bugs: the percentage of tasks where every potential bug the agent reports is backed by a valid fail-to-pass test, building on test validity (arXiv 2608.04682v1). 0-100, higher is better. Every model ran in the Claude Code harness; from the project site's data/site-data.json (full 400-task main set).
Metric: Potential Bugs Revealed (%). Source: hbinli.github.io. Saturation forecast: Around February 2027. 18 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.8 | 75 | #39 |
| 2 | GLM-5.2 | 69 | #80 |
| 3 | GPT-5.5 | 65.5 | #33 |
| 4 | Qwen 3.7 Max | 59.8 | #85 |
| 5 | Hy3 | 58.5 | #100 |
| 6 | Gemini 3.1 Pro (Preview) | 57 | #64 |
| 7 | Ring-2.6-1T | 56.3 | #176 |
| 8 | Claude Sonnet 4.6 | 53.8 | #94 |
| 9 | DeepSeek V4 Pro | 53.3 | #112 |
| 10 | Kimi K2.6 | 51.8 | #113 |
| 11 | Kimi K2.7 Code | 47.8 | #95 |
| 12 | Qwen 3.5 397B A17B | 42 | #154 |
| 13 | Qwen 3.5 122B A10B | 40 | #184 |
| 14 | Ling-3.0-flash | 38.75 | #215 |
| 15 | MiniMax-M2.7 | 38.5 | #270 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=active-swe-revealed · How It Works · Data refreshed daily, snapshot 2026-11-05.