ICE-Guard - Framing: leaderboard
Share of decision pairs that flip when the same facts are restated positively or negatively (for example 95% survival versus 5% mortality). Unlike the demographic slice, the swap concerns wording, not the person decided about.
Metric: Decision flip rate under an irrelevant swap (%, lower is better). Source: arxiv.org. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GLM-5 | 0 | #260 |
| 2 | GPT-OSS-120B | 1 | #700 |
| 3 | Claude Opus 4.6 | 2 | #102 |
| 4 | GPT-5.2 | 3 | #176 |
| 5 | Claude Sonnet 4.5 | 4 | #250 |
| 6 | Kimi K2.5 | 6.2 | #256 |
| 7 | Gemini 3 Flash | 7 | #164 |
| 8 | DeepSeek V3.2 | 7 | #406 |
| 9 | Gemini 2.5 Pro | 8 | #266 |
| 10 | Qwen 3 32B | 10 | #960 |
Interactive version: theaggregate.ai/benchmark?slug=ice-guard-framing · How It Works · Data refreshed daily, snapshot 2026-11-13.