RULER: leaderboard
13 synthetic long-context tasks from NVIDIA (2024) in four groups (needle-in-a-haystack retrieval, variable tracing, word-frequency aggregation, distractor-padded QA) generated at 4K to 128K tokens.
Metric: Avg Accuracy (%). Source: arxiv.org. Source: github.com. Saturation forecast: Estimated already saturated. 45 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Jamba 1.5 Large | 96 | #669 |
| 2 | Gemini 1.5 Pro | 95.8 | #445 |
| 3 | Qwen2.5-14B-Instruct-1M | 95.7 | #578 |
| 4 | Qwen 3 235B A22B | 95 | #313 |
| 5 | Qwen 3 14B | 94.6 | #537 |
| 6 | Jamba 1.5 Mini | 93.9 | #983 |
| 7 | Qwen 3 32B | 93.7 | #443 |
| 8 | Qwen2.5-7B-Instruct-1M | 91.8 | #894 |
| 9 | Qwen 3 30B A3B | 91.6 | #493 |
| 10 | GPT-4 Preview (1106) | 91.6 | #418 |
| 11 | glm-4-9B | 89.9 | #889 |
| 12 | Llama 3.1 70B | 89.6 | #592 |
| 13 | Qwen 3 8B | 89.1 | #674 |
| 14 | Llama 3.1 8B | 88.3 | #1151 |
| 15 | Yi Chat 34B | 87.5 | #976 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=ruler · How It Works · Data refreshed daily, snapshot 2026-10-20.