Diplomacy: Steerability: leaderboard
Prompt responsiveness difference, not absolute playing strength. Signed native units, no established finite domain.
Metric: Mean game-score difference: optimized aggressive prompt minus baseline prompt. Source: goodstartlabs.com. 22 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GLM-4.6 | 32.5 | #252 |
| 2 | GPT-5 (Minimal) | 21.9 | #98 (GPT-5) |
| 3 | O3 | 17.17 | #124 |
| 4 | GPT-5 Mini | 12.02 | #186 |
| 5 | Llama 4 Maverick | 9.7 | #464 |
| 6 | Grok 4 Fast | 9.7 | #254 |
| 7 | GPT-5 (Medium) | 8.95 | #98 (GPT-5) |
| 8 | GPT-OSS-120B | 8.04 | #341 |
| 9 | Hermes 4 405B | 6.35 | #334 |
| 10 | Kimi K2 | 5.85 | #243 |
| 11 | O4 Mini | 5.32 | #180 |
| 12 | Gemini 3 Pro | 5.3 | #80 |
| 13 | Grok 4 Fast (Reasoning) | 3.1 | #254 (Grok 4 Fast) |
| 14 | Gemini 2.5 Pro | 1.85 | #152 |
| 15 | Claude Opus 4.5 | 1.6 | #83 |
Interactive version: theaggregate.ai/benchmark?slug=diplomacy-steerability · How It Works · Data refreshed daily, snapshot 2026-10-20.