NoLiMa: leaderboard
Long-context benchmark that goes beyond literal needle matching. Tests whether models can truly understand and retrieve information from long documents, not just pattern-match on exact strings.
Metric: Base Score (%). Source: github.com. Saturation forecast: Estimated already saturated. 22 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o | 99.3 | #343 |
| 2 | Llama 3.3 70B | 97.3 | #579 |
| 3 | GPT-4.1 | 97 | #248 |
| 4 | Llama 3.1 405B | 94.7 | #466 |
| 5 | Llama 3.1 70B | 94.5 | #592 |
| 6 | Gemini 1.5 Pro | 92.6 | #445 |
| 7 | Jamba 1.5 Mini | 92.4 | #983 |
| 8 | Command-R+ | 90.9 | #891 |
| 9 | Llama 4 Maverick | 90.1 | #464 |
| 10 | Gemini 2.0 Flash | 89.4 | #342 |
| 11 | Gemma 3 27B | 88.6 | #607 |
| 12 | Mistral Large 2 (Jul) | 87.9 | #484 |
| 13 | Claude 3.5 Sonnet | 87.6 | #350 |
| 14 | Gemma 3 12B | 87.4 | #671 |
| 15 | GPT-4o Mini | 84.9 | #599 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=nolima · How It Works · Data refreshed daily, snapshot 2026-10-20.