ATLAS: leaderboard
Long-context language models now advertise context windows up to millions of tokens, yet evaluations typically report a single length or a narrow task family, masking two failure modes: performance can collapse as length grows, and strong retrieval need not transfer to downstream use. We present ATLAS, a benchmarking framework that redefines long-context evaluation as length-dependent capability p
Metric: ATLAScore (self-reported). Source: benchmarklist.com. Saturation forecast: Around July 2027. 23 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.6 (Max) | 70.55 | #63 (Claude Opus 4.6) |
| 2 | Gemini 3.1 Pro (Preview) (High) | 68.52 | #58 (Gemini 3.1 Pro (Preview)) |
| 3 | GPT-5.5 (xHigh) | 67.77 | #31 (GPT-5.5) |
| 4 | GPT-5.4 (xHigh) | 63.23 | #79 (GPT-5.4) |
| 5 | Gemini 3 Flash (Preview) (High) | 63.13 | #82 (Gemini 3 Flash (Preview)) |
| 6 | Gemini 3 Pro (Preview) (High) | 62.6 | #66 (Gemini 3 Pro (Preview)) |
| 7 | DeepSeek V4 Pro (Max) | 59.11 | #102 (DeepSeek V4 Pro) |
| 8 | GPT-5.2 (xHigh) | 58.2 | #110 (GPT-5.2) |
| 9 | Qwen 3.5 397B A17B (Reasoning) | 54.43 | #146 (Qwen 3.5 397B A17B) |
| 10 | Claude Opus 4.5 (High) | 51.85 | #83 (Claude Opus 4.5) |
| 11 | Kimi K2.6 (Thinking) | 50.47 | #104 (Kimi K2.6) |
| 12 | Kimi K2.5 (Thinking) | 47.11 | #144 (Kimi K2.5) |
| 13 | GLM-5.1 (Thinking) | 46.35 | #97 (GLM-5.1) |
| 14 | GLM-5 (Thinking) | 44.28 | #141 (GLM-5) |
| 15 | Qwen 3 235B A22B 2507 Instruct | 35.04 | #300 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=atlas · How It Works · Data refreshed daily, snapshot 2026-10-20.