LongBench v2
A long-context benchmark that measures whether models can actually use extended context windows for reasoning and retrieval.
Top models on LongBench v2 — October 7, 2026
As of October 7, 2026, Qwen3.8 Max leads the LongBench v2 leaderboard with 66.3% , followed by Beam (65.5%) and Claude Opus 4.5 (64.4%).
Qwen3.8 Max
Alibaba
Beam
Reflection AI
Claude Opus 4.5
Anthropic
17 modelsReasoning25% of category scoreCurrentUpdated October 7, 2026
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | Qwen3.8 MaxAlibaba | 66.3% | Not reported | Open |
| 2 | BeamReflection AI | 65.5% | Not reported | Pending |
| 3 | Claude Opus 4.5Anthropic | 64.4% | Not reported | Closed |
| 4 | Qwen3.5 397BAlibaba | 63.2% | Not reported | Open |
| 5 | Qwen3.6 PlusAlibaba | 62% | Not reported | Closed |
| 6 | Nemotron 3 UltraNVIDIA | 61.9% | Not reported | Open |
| 7 | Kimi K2.5Moonshot AI | 61% | Not reported | Open |
| 8 | GLM-5Z.AI | 60.8% | Not reported | Open |
| 9 | Qwen3.5-27BAlibaba | 60.6% | Not reported | Open |
| 10 | Qwen3.5-122B-A10BAlibaba | 60.2% | Not reported | Open |
| 11 | Agents-A1InternScience | 60.2% | Not reported | Open |
| 12 | Qwen3.5-35B-A3BAlibaba | 59% | Not reported | Open |
| 13 | Agents-A1-4BInternScience | 52.1% | Not reported | Open |
| 14 | DeepSeek V4 Pro BaseDeepSeek | 51.5% | Not reported | Open |
| 15 | DeepSeek V4 Flash BaseDeepSeek | 44.7% | Not reported | Open |
| 16 | MiniCPM5-2BOpenBMB | 43.7% | Not reported | Open |
| 17 | LLaDA2.2-miniInclusionAI | 35.0% | Not reported | Open |
According to BenchLM.ai, Qwen3.8 Max leads the LongBench v2 benchmark with a score of 66.3%, followed by Beam (65.5%) and Claude Opus 4.5 (64.4%). The top models are clustered within 1.9 points, suggesting this benchmark is nearing saturation for frontier models.
17 models have been evaluated on LongBench v2. The benchmark falls in the Reasoning category. The Reasoning leaderboard ranks models by a weighted category score, and LongBench v2 contributes 25% of it. BenchAlign v5.8 also uses it in the overall ranking.
About LongBench v2
Year
2025
Tasks
Long-context tasks
Format
Extended-context retrieval and reasoning
Difficulty
Hard long-context
LongBench v2 is useful because context-window size alone is not a capability. It measures whether a model can retain, retrieve, and reason over long inputs effectively.
Freshness and provenance
Version
LongBench v2 2025
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does LongBench v2 measure?
A long-context benchmark that measures whether models can actually use extended context windows for reasoning and retrieval.
Which model scores highest on LongBench v2?
Qwen3.8 Max by Alibaba currently leads with a score of 66.3% on LongBench v2.
How many models are evaluated on LongBench v2?
17 AI models have published results on LongBench v2 in the BenchLM catalog.
Compare top models on LongBench v2
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 5,500+ readers.
One email each week. Unsubscribe anytime.