LIBRA - PasskeyWithLibrusec: leaderboard
LIBRA long-context subtask: retrieve hidden passkeys when distractor text from the Librusec corpus is mixed into the context.
Metric: Dataset Total Score (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Llama 3.1 8B Instruct | 100 | #1035 |
| 2 | Qwen 3 30B A3B 2507 Instruct | 100 | #488 |
| 3 | Phi-3.5-mini-instruct | 91.25 | #1167 |
| 4 | Qwen 3.5 35B A3B | 84.72 | #256 |
| 5 | Qwen 3.5 9B | 80.6 | #379 |
| 6 | Qwen 3.5 122B A10B | 80.25 | #173 |
| 7 | Qwen 3.5 4B | 75.93 | #475 |
| 8 | Qwen 3.5 2B | 72.64 | #1148 |
| 9 | Qwen 3.5 0.8B | 69.03 | #1424 |
| 10 | LFM2.5-1.2B Instruct | 66.08 | #1462 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=libra-passkeywithlibrusec · How It Works · Data refreshed daily, snapshot 2026-10-20.