TRACE Bench - Short-Term Memory: leaderboard
Metric: Memory-probe completion rate (%). Source: arxiv.org. 26 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.5 | 92 |
| 2 | MiMo-V2.5 | 91 |
| 3 | Gemma 4 31B (Non-reasoning) | 90.5 |
| 4 | MiMo-V2.5-Pro | 90 |
| 5 | MiniMax-M2.7 | 88.5 |
| 6 | Kimi K2.5 (Non-reasoning) | 87.5 |
| 7 | DeepSeek V4 Flash | 87 |
| 8 | GLM-5 (Non-reasoning) | 87 |
| 9 | Claude Sonnet 4.6 | 86 |
| 10 | Qwen 3 8B (Non-reasoning) | 85.5 |
| 11 | L3-8B-Stheno-v3.2 | 85.43 |
| 12 | Qwen 3.5 27B (Non-reasoning) | 85 |
| 13 | DeepSeek V4 Pro (Non-reasoning) | 85 |
| 14 | Mistral Small 3.2 | 84.5 |
| 15 | Qwen 3 14B (Non-reasoning) | 84 |
Interactive version: theaggregate.ai/benchmark?slug=trace-bench-short-term-memory · How It Works · Data refreshed daily, snapshot 2026-09-19.