Syco-bench - Mirroring (Provider Web System Prompt): leaderboard
The model gives its view after the user takes one stance and then the opposite one, with the system prompt of the provider's chat interface; the number is the mean absolute shift between the two judged positions (0-10). Measures stance mirroring, not factual capitulation.
Metric: Mean absolute shift in the model's stance between the two opposite user stances (-5..5 scale, so 0-10). Source: www.syco-bench.com. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude 3.7 Sonnet (Thinking) | 1.02 | #458 (Claude 3.7 Sonnet) |
| 2 | Gemini 2.5 Pro (Preview) | 1.02 | #253 |
| 3 | Gemini 2.5 Flash (Preview) | 1.4 | #785 |
| 4 | Claude 3.7 Sonnet | 1.48 | #458 |
| 5 | Grok 3 Mini Beta | 1.5 | #616 |
| 6 | Claude 3 Opus | 1.93 | #1455 |
| 7 | O4 Mini | 2 | #353 |
| 8 | O3 | 2.28 | #228 |
| 9 | Grok 3 | 2.53 | #602 |
| 10 | Claude 3.5 Haiku | 2.6 | #1553 |
| 11 | GPT-4o Mini | 2.83 | #1623 |
| 12 | GPT-4o ChatGPT | 3.18 | #466 |
Interactive version: theaggregate.ai/benchmark?slug=syco-bench-mirroring-provider-web-system-prompt · How It Works · Data refreshed daily, snapshot 2026-11-13.