LIBRA - MatreshkaNames *: leaderboard
LIBRA updated long-context subtask: recover nested names from Matreshka-style contexts, testing precise entity retrieval over very long inputs.
Metric: Dataset Total Score (%). Source: huggingface.co. Saturation forecast: Around January 2027. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 30B A3B 2507 Instruct | 81.2 | #488 |
| 2 | Qwen 3.5 35B A3B | 68.97 | #256 |
| 3 | Llama 3.1 8B Instruct | 68.02 | #1035 |
| 4 | Qwen 3.5 122B A10B | 67.39 | #173 |
| 5 | Qwen 3.5 9B | 65.44 | #379 |
| 6 | Qwen 3.5 4B | 61.66 | #475 |
| 7 | Qwen 3.5 2B | 58.98 | #1148 |
| 8 | Qwen 3.5 0.8B | 56.05 | #1424 |
| 9 | Phi-3.5-mini-instruct | 37.4 | #1167 |
| 10 | LFM2.5-1.2B Instruct | 6.72 | #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-matreshkanames-star · How It Works · Data refreshed daily, snapshot 2026-10-20.