Altered Riddles: leaderboard
Reasoning benchmark for conditioned override, testing whether models fall back to memorized answers when familiar riddles are deliberately modified with constraints, context swaps, meaning shifts, or bias probes.
Metric: Conditioned Override Rate. Source: huggingface.co. Source: benchmarklist.com. Saturation forecast: Around August 2027. 23 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 (High) | 0.31 | #79 (GPT-5.4) |
| 2 | GPT-5.4 Mini | 0.4 | #212 |
| 3 | Kimi K2.5 (High) | 0.43 | #144 (Kimi K2.5) |
| 4 | Kimi K2.6 (High) | 0.44 | #104 (Kimi K2.6) |
| 5 | Gemma 4 31B (IT) | 0.46 | #176 |
| 6 | Mistral Small 4 | 0.53 | #517 |
| 7 | Kimi K2.5 | 0.54 | #144 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=altered-riddles · How It Works · Data refreshed daily, snapshot 2026-10-20.