SteerBench-Work: leaderboard
Metric: Mean Trial Accuracy (%). Source: steerbench.com. 30 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Flash Lite (Minimal) | 92.8 |
| 2 | GPT-5.4 Mini (High) | 91.9 |
| 3 | GPT-5.4 Mini (Non-reasoning) | 91.9 |
| 4 | DeepSeek V4 Pro (Non-reasoning) | 91.7 |
| 5 | Gemini 3.5 Flash (Minimal) | 90.6 |
| 6 | Kimi K2.6 (Non-reasoning) | 90.4 |
| 7 | DeepSeek V4 Flash (Reasoning) | 90 |
| 8 | Kimi K2.6 (Thinking) | 88.9 |
| 9 | DeepSeek V4 Flash (Non-reasoning) | 88.9 |
| 10 | DeepSeek V4 Pro (Reasoning) | 87.5 |
| 11 | Claude Opus 4.8 (High) | 87.4 |
| 12 | GPT-OSS-20B (Low) | 86.2 |
| 13 | GPT-OSS-20B (High) | 85.5 |
| 14 | Claude Haiku 4.5 (Non-reasoning) | 85.3 |
| 15 | Claude Haiku 4.5 (High) | 85.1 |
Interactive version: theaggregate.ai/benchmark?slug=steerbench-work · How It Works · Data refreshed daily, snapshot 2026-09-19.