HELMET (128K): leaderboard
HELMET — Princeton long-context benchmark evaluated at 128K context: RAG, passage re-ranking, many-shot ICL, summarization and long-document QA.
Metric: Average Score. Source: princeton-nlp.github.io. Saturation forecast: Estimated already saturated. 64 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o (2024-08-06) | 64.8 | #336 |
| 2 | Gemini 1.5 Pro | 64.4 | #445 |
| 3 | GPT-4o Mini | 55.7 | #599 |
| 4 | Gemini 1.5 Flash | 52.1 | #618 |
| 5 | Llama 3.1 70B Instruct | 49.7 | #561 |
| 6 | Llama 3.3 70B Instruct | 48.2 | #534 |
| 7 | Jamba 1.5 Mini | 46.9 | #983 |
| 8 | Llama 3.1 8B Instruct | 46.5 | #1035 |
| 9 | Gemma 3 27B | 44.5 | #607 |
| 10 | Gemma 3 27B (IT) | 41.9 | #523 |
| 11 | Llama 3.1 70B | 41.7 | #592 |
| 12 | Gemma 3 12B | 40.6 | #671 |
| 13 | Yi-34B-200K | 40.3 | #1176 |
| 14 | GPT-4 | 39.2 | #438 |
| 15 | Claude 3.5 Sonnet | 38.4 | #350 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=helmet-128k · How It Works · Data refreshed daily, snapshot 2026-10-20.