StaminaBench (Mini-SWE): leaderboard
StaminaBench stress-tests coding agents over up to 101 consecutive change requests, measuring how many turns an agent handles before its first failure on procedurally generated REST-API scenarios with a two-retry feedback loop. This board reports Mini-SWE-Agent harness: average consecutive interaction turns passed before the first failing change request, over 20 procedurally generated REST-API scenarios of up to 101 turns with two retries; higher is better.
Metric: Mini-SWE-Agent harness: average consecutive interaction turns passed before the first failing change request, over 20 procedurally generated REST-API scenarios of up to 101 turns with two retries; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.5 122B A10B | 33.1 | #331 |
| 2 | GLM-5 | 15.1 | #260 |
| 3 | Kimi K2.5 | 14.9 | #256 |
| 4 | Qwen3 Coder Next | 6.2 | #714 |
| 5 | Devstral 2 | 5 | #1003 |
| 6 | Devstral Small 2 | 2.3 | #1345 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=staminabench-mini-swe · How It Works · Data refreshed daily, snapshot 2026-11-13.