Harbor Adapters - SWE-bench Multilingual (Terminus-2): leaderboard
Metric: Resolved rate (%; 50 tasks, mean of 3 trials). Source: arxiv.org. Saturation forecast: Around December 2026. 8 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.6 | 80 |
| 2 | Gemini 3.1 Pro (Preview) | 76.7 |
| 3 | Gemini 3 Flash (Preview) | 74.7 |
| 4 | Claude Sonnet 4.6 | 73.3 |
| 5 | Claude Haiku 4.5 | 68.7 |
| 6 | GPT-5.4 (Medium) | 60.7 |
| 7 | GPT-5 Mini | 21.3 |
| 8 | GPT-5 Nano | 4.7 |
Interactive version: theaggregate.ai/benchmark?slug=harbor-adapters-swe-bench-multilingual-terminus-2 · How It Works · Data refreshed daily, snapshot 2026-09-25.