Reasoning benchmark
LongBench v2 leaderboard
Every model the catalog carries a published LongBench v2 value for, ranked by that value.
CategoryReasoning
MeasureExtended-context retrieval and reasoning
TasksLong-context tasks
DifficultyHard long-context
Published byLongBench v2
A long-context benchmark that measures whether models can actually use extended context windows for reasoning and retrieval.
LongBench v2 ranking
14 models with a published LongBench v2 value, ordered by that value, highest first. Models the source has not scored on this benchmark are not listed — they are unmeasured, not last.
| Rank | Model | Provider | Extended-context retrieval and reasoning |
|---|---|---|---|
| 1 |  Qwen3.8 Max | Alibaba | 66.3 |
| 2 |  Claude Opus 4.5 | Anthropic | 64.4 |
| 3 | Qwen3.5 397B | Alibaba | 63.2 |
| 4 | Qwen3.6 Plus | Alibaba | 62 |
| 5 |  Nemotron 3 Ultra | NVIDIA | 61.9 |
| 6 |  Kimi K2.5 | Moonshot AI | 61 |
| 7 |  GLM-5 | Z.AI | 60.8 |
| 8 | Qwen3.5-27B | Alibaba | 60.6 |
| 9 |  Agents-A1 | InternScience | 60.2 |
| 9 | Qwen3.5-122B-A10B | Alibaba | 60.2 |
| 11 | Qwen3.5-35B-A3B | Alibaba | 59 |
| 12 | Agents-A1-4B | InternScience | 52.1 |
| 13 | OPMiniCPM5-2B | OpenBMB | 43.7 |
| 14 |  LLaDA2.2-mini | InclusionAI | 34.99 |
Evidence key: Observed
Rows are ordered by the value LongBench v2 published, highest first. The source does not state whether a higher value is the better result, so this page does not either: for a benchmark that measures a rate of failure, read the table from the bottom.