AthenaBench - TAA: leaderboard
Name the threat actor behind 100 anonymised incident reports; exact-actor accuracy (alias matches count, related groups do not).
Metric: Accuracy (%) attributing anonymised incident reports to a threat actor (100 reports). Source: raw.githubusercontent.com. Saturation forecast: Around 2029. 9 models tracked. Updated 17 Nov 2026.
Top models
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | Gemini 2.5 Pro | 31 | #274 | |
| 2 | Gemini 2.5 Flash | 30 | #467 | |
| 3 | Llama 3.3 70B Instruct | 26 | #1387 | |
| 4 | Llama 3.1 8B Instruct | 24 | #4576 | |
| 5 | Llama 3 70B Instruct | 22 | #1689 | |
| 6 | Qwen 3 14B | Reasoning | 17 | #1264 |
| 7 | Llama-Primus-Merged | 17 | #2461 | |
| 8 | Qwen 3 8B | Thinking | 16 | #1868 |
| 9 | Qwen 3 4B 2507 Instruct | 15 | #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-taa · How It Works · Data refreshed daily, snapshot 2026-11-17.