Harbor Adapters - BIX-Bench (Terminus-2): leaderboard
Metric: Accuracy (%; 50 questions, LLM judge, mean of 3 trials). Source: arxiv.org. Saturation forecast: Around May 2027. 8 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 43.3 |
| 2 | Gemini 3 Flash (Preview) | 39.3 |
| 3 | Claude Opus 4.6 | 30 |
| 4 | GPT-5.4 (Medium) | 30 |
| 5 | Claude Sonnet 4.6 | 29.3 |
| 6 | GPT-5 Mini | 19.3 |
| 7 | Claude Haiku 4.5 | 15.3 |
| 8 | GPT-5 Nano | 7.3 |
Interactive version: theaggregate.ai/benchmark?slug=harbor-adapters-bix-bench-terminus-2 · How It Works · Data refreshed daily, snapshot 2026-09-25.