ImplicitMemBench: 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 the overall score, the mean of procedural-memory accuracy, classical-conditioning accuracy and the judged priming score; higher is better.

Metric: Overall score: mean of procedural-memory accuracy, classical-conditioning accuracy (%) and the GPT-4o-mini-judged priming influence score (0 to 95 bands) on the 300 ImplicitMemBench items (learning, interference and test phases, first-attempt scoring, mean of three runs at temperature 0); higher is better. Source: arxiv.org. Saturation forecast: Around August 2028. 17 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1DeepSeek R165.3#482
2Qwen 3 32B64.13#972
3GPT-563#178
4Qwen 3 8B62.35#1873
5O361.79#237
6O4 MiniHigh60.87#368
7GLM-4.558.59#542
8Gemini 2.5 Pro55.69#276
9Claude Opus 4.155.65#278
10Gemini 2.5 Flash55.43#463
11GPT-4o Mini50.88#1667
12Qwen 2.5 72B Instruct50.78#1117
13GPT-4o50.32#751
14Claude Sonnet 449.84#410
15Llama 3.3 70B Instruct49.44#1385

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

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