needle-1M-bench: leaderboard
Long-context needle-retrieval benchmark over dense scientific-paper text, scaling haystacks up to one million tokens and reporting recall.
Metric: Overall recall (%). Source: huggingface.co. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 100 | #152 |
| 2 | DeepSeek V4 Pro | 100 | #102 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=needle-1m-bench · How It Works · Data refreshed daily, snapshot 2026-10-20.