ImplicitMemBench - Procedural Memory: leaderboard

ImplicitMemBench (2026) measures implicit memory in LLMs, whether behavior adapts without explicit recall after learning, interference and test phases in a conversation. This board is first-attempt accuracy in percent at applying a procedure learned earlier in the conversation after interference; higher is better.

Metric: First-attempt accuracy (%) on the 100 procedural-memory items: a one-shot procedure learned in conversation must be applied after interference, checked by rules or GPT-4o-mini (learning, interference and test phases, first-attempt scoring, mean of three runs at temperature 0); higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 17 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Claude Opus 4.176.67#278
2DeepSeek R176.33#482
3GLM-4.576.33#542
4O376#237
5Qwen 3 32B75.67#972
6GPT-575.33#178
7Qwen 3 8B75.33#1873
8Gemini 2.5 Pro74.33#276
9Gemini 2.5 Flash72.33#463
10O4 MiniHigh70.67#368
11GPT-4o61.67#751
12GPT-4o Mini61.67#1667
13Qwen 2.5 72B Instruct61#1117
14Llama 3.3 70B Instruct58.33#1385
15Claude Sonnet 451.67#410

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=implicitmembench-procedural-memory · How It Works · Data refreshed daily, snapshot 2026-11-21.