HSE SWE-bench Verified - Frozen Transfer Pi: leaderboard
HSE v1 reports resolved rates on 500 SWE-bench Verified tasks and 113 DeepSWE tasks, using 3 runs. The default model is GLM-5.2 in the Pi or Claude Code harness. Its static library of 10 skills is evolved on SWE-Gym, with 8 success skills and 2 recovery skills. Gates 1–8 explicitly adapt to the target benchmark and remain distinct from frozen transfer. Cross-model GPT 5.5 and Opus 4.8 evaluations use Pi only. Model, harness, skill library, training scope, and gate identities are preserved. Metric: Resolved rate (%).
Metric: Resolved rate (%). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | HSE study GLM-5.2 Pi HSE SWE-Gym library | 75.4 | |
| 2 | HSE study GLM-5.2 Pi SWESkillsBench | 74.7 | |
| 3 | HSE study GLM-5.2 Pi GEPA | 74.6 | |
| 4 | HSE study GLM-5.2 Pi No Skills | 73.5 | |
| 5 | HSE study GLM-5.2 Pi ReasoningBank | 72.8 |
Interactive version: theaggregate.ai/benchmark?slug=hse-swe-bench-verified-frozen-transfer-pi · How It Works · Data refreshed daily, snapshot 2026-10-28.