MemTrace - Conflict Resolution: leaderboard
False-premise questions over a long multi-session history read directly in the model's context: the question asserts something that contradicts a remembered user fact, and the answer must correct it; MemTrace (Long et al., 2026). Gist accuracy in percent from a GPT-4o judge at the saturated checkpoints; higher is better.
Metric: Gist accuracy (%; GPT-4o judge; the model answers from the visible multi-session history in its context; conflict probes at checkpoints W7-W8: the question asserts a false premise that contradicts a remembered fact, and the answer must correct it). Source: arxiv.org. Saturation forecast: Around February 2027. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.5 35B A3B | 88.6 | #480 |
| 2 | Gemini 3 Flash | 79.8 | #159 |
| 3 | GPT-5 Nano | 79 | #888 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=memtrace-conflict-resolution · How It Works · Data refreshed daily, snapshot 2026-11-12.