Managing Procedural Memory in LLM Agents: leaderboard
Procedural memory is increasingly used to improve LLM agents on recurring workplace tasks, yet its ability to produce reusable skills remains poorly understood. We introduce AFTER, a benchmark of 382 realistic enterprise tasks spanning six professional roles and 22 procedural skills, designed to evaluate how skills transfer across tasks, roles, and model backbones. The benchmark includes controlle
Metric: Aggregate M2 task success (%), best skill setting (self-reported). Source: benchmarklist.com. Saturation forecast: Around October 2028. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 | 50.1 | #79 |
| 2 | GPT-OSS-120B | 45.6 | #341 |
| 3 | GPT-5.4 Mini | 44.9 | #212 |
| 4 | Gemma 4 31B | 41.3 | #200 |
| 5 | Qwen 3.5 122B A10B | 39.7 | #173 |
| 6 | Gemma 4 26B A4B | 39.7 | #358 |
| 7 | DeepSeek V4 Flash | 37.1 | #137 |
| 8 | Qwen 3.5 35B A3B | 32.2 | #256 |
| 9 | GPT-OSS-20B | 31.3 | #518 |
| 10 | Gemma 4 E4B | 24 | #767 |
| 11 | Qwen 3.5 9B | 18.7 | #379 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=managing-procedural-memory-in-llm-agents · How It Works · Data refreshed daily, snapshot 2026-10-20.