MemDrift: leaderboard
Metric: Deflection score under biased memory (1-5; Claude Opus 4.6 judge compares five tool calls against the no-memory calls, 1 = unchanged), direct memory injection, mean of five bias dimensions (105 scenarios). Source: arxiv.org. Saturation forecast: Around July 2027. 7 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.2 (Medium) | 3.53 |
| 2 | GPT-5.4 (Medium) | 3.79 |
| 3 | Qwen 3.5 397B A17B (Thinking) | 4.42 |
| 4 | Gemini 2.5 Pro (Medium) | 4.56 |
| 5 | Kimi K2.5 (Thinking) | 4.62 |
| 6 | Gemini 3.1 Pro (Preview) (Medium) | 4.67 |
| 7 | Claude Sonnet 4.5 | 4.79 |
Interactive version: theaggregate.ai/benchmark?slug=memdrift · How It Works · Data refreshed daily, snapshot 2026-09-26.