AthenaBench - ATE: leaderboard
Name the MITRE ATT&CK technique a threat-report passage describes, 500 items; accuracy on the top-level technique ID (sub-technique suffix ignored).
Metric: Accuracy (%) extracting ATT&CK techniques from threat reports (500 items). Source: raw.githubusercontent.com. Saturation forecast: Estimated already saturated. 9 models tracked. Updated 17 Nov 2026.
Top models
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | Gemini 2.5 Pro | 76.2 | #274 | |
| 2 | Gemini 2.5 Flash | 51.6 | #467 | |
| 3 | Llama-Primus-Merged | 33.8 | #2461 | |
| 4 | Llama 3 70B Instruct | 31.6 | #1689 | |
| 5 | Llama 3.3 70B Instruct | 30.4 | #1387 | |
| 6 | Qwen 3 14B | Reasoning | 19.4 | #1264 |
| 7 | Llama 3.1 8B Instruct | 16.4 | #4576 | |
| 8 | Qwen 3 8B | Thinking | 11.8 | #1868 |
| 9 | Qwen 3 4B 2507 Instruct | 5.6 | #2215 |
No result here: #1 Claude Fable 5.2, #2 GPT-6.2 Astra, #3 Gemini 4 Carbon.
Interactive version: theaggregate.ai/benchmark?slug=athenabench-ate · How It Works · Data refreshed daily, snapshot 2026-11-17.