ImplicitMemBench - Procedural Memory: leaderboard
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 January 2027. 17 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.1 | 76.67 |
| 2 | DeepSeek R1 | 76.33 |
| 3 | GLM-4.5 | 76.33 |
| 4 | O3 | 76 |
| 5 | Qwen 3 32B | 75.67 |
| 6 | GPT-5 | 75.33 |
| 7 | Qwen 3 8B | 75.33 |
| 8 | Gemini 2.5 Pro | 74.33 |
| 9 | Gemini 2.5 Flash | 72.33 |
| 10 | O4 Mini (High) | 70.67 |
| 11 | GPT-4o | 61.67 |
| 12 | GPT-4o Mini | 61.67 |
| 13 | Qwen 2.5 72B Instruct | 61 |
| 14 | Llama 3.3 70B Instruct | 58.33 |
| 15 | Claude Sonnet 4 | 51.67 |
Interactive version: theaggregate.ai/benchmark?slug=implicitmembench-procedural-memory · How It Works · Data refreshed daily, snapshot 2026-10-07.