SteerBench-Work - Pass^5: leaderboard
Metric: Scenarios Correct in All 5 Trials (%). Source: steerbench.com. 30 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Flash Lite (Minimal) | 91.5 |
| 2 | GPT-5.4 Mini (Non-reasoning) | 89.6 |
| 3 | Gemini 3.5 Flash (Minimal) | 89.6 |
| 4 | GPT-5.4 Mini (High) | 87.7 |
| 5 | Claude Opus 4.8 (High) | 85.8 |
| 6 | Kimi K2.6 (Non-reasoning) | 84.9 |
| 7 | Claude Haiku 4.5 (Non-reasoning) | 84.9 |
| 8 | Kimi K2.6 (Thinking) | 83 |
| 9 | DeepSeek V4 Pro (Reasoning) | 83 |
| 10 | Claude Sonnet 4.6 (Non-reasoning) | 82.1 |
| 11 | DeepSeek V4 Flash (Reasoning) | 82.1 |
| 12 | GPT-5.4 (Non-reasoning) | 81.1 |
| 13 | Claude Haiku 4.5 (High) | 81.1 |
| 14 | GPT-5.5 (Non-reasoning) | 80.2 |
| 15 | DeepSeek V4 Pro (Non-reasoning) | 79.2 |
Interactive version: theaggregate.ai/benchmark?slug=steerbench-work-pass-pow-5 · How It Works · Data refreshed daily, snapshot 2026-09-19.