AMA-Bench - Software: leaderboard
AMA-Bench software slice measures agent memory on software-engineering trajectories.
Metric: Average Score (%). Source: huggingface.co. Saturation forecast: Around June 2027. 26 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 Mini | 57.33 | #186 |
| 2 | GPT-5.2 | 54.1 | #110 |
| 3 | Qwen 3 32B | 48.47 | #443 |
| 4 | Qwen2.5-14B-Instruct-1M | 43.72 | #578 |
| 5 | Qwen 3 14B | 39.6 | #537 |
| 6 | Qwen 3 8B | 38.42 | #674 |
| 7 | Gemini 2.5 Flash | 38.38 | #245 |
| 8 | Claude 3.5 Haiku | 25.07 | #568 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=ama-bench-software · How It Works · Data refreshed daily, snapshot 2026-10-20.