LLM Stats (LongBench v2): leaderboard
6 task categories testing long-context understanding at 8K–128K tokens, including single/multi-document QA, summarization, few-shot learning, and code completion.
Metric: Score (%). Source: llm-stats.com. Saturation forecast: Around July 2027. 18 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.8 Max | 66.3 | #35 |
| 2 | Qwen 3.5 397B A17B | 63.2 | #146 |
| 3 | Qwen 3.6 Plus | 62 | #120 |
| 4 | Nemotron 3 Ultra | 61.9 | #235 |
| 5 | MiniMax M1 80k | 61.5 | #432 |
| 6 | Kimi K2.5 | 61 | #144 |
| 7 | MiniMax M1 40k | 61 | #683 |
| 8 | Qwen 3.5 27B | 60.6 | #187 |
| 9 | MiMo-V2-Flash | 60.6 | #372 |
| 10 | Qwen 3 Max (Thinking) | 60.6 | #208 (Qwen 3 Max) |
| 11 | Qwen 3.5 122B A10B | 60.2 | #173 |
| 12 | Qwen 3.5 35B A3B | 59 | #256 |
| 13 | Qwen 3.5 9B | 55.2 | #379 |
| 14 | Qwen 3.5 4B | 50 | #475 |
| 15 | DeepSeek V3 | 48.7 | #321 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=llm-stats-longbench-v2 · How It Works · Data refreshed daily, snapshot 2026-10-20.