Inference
LongBench v2 authors

LongBench v2

A long-context benchmark that measures whether models can actually use extended context windows for reasoning and retrieval.

ModelsModels with a result.14
Top resultBest result on this test.N/A
Top-3 spreadPoints from first to third.N/A
YearYear of release.2025

Result and price

020406080100$1$3$10$30RESULTOUTPUT PRICE PER 1M TOKENS · LOG SCALE

Results

14 results
#ModelResultThe score from the source.
—Qwen3.8 Max66.3%
—Claude Opus 4.564.4%
—Qwen3.5 397B63.2%
—Qwen3.6 Plus62.0%
—Nemotron 3 Ultra61.9%
—Kimi K2.561.0%
—GLM-560.8%
—Qwen3.5-27B60.6%
—Qwen3.5-122B-A10B60.2%
—Agents-A160.2%
—Qwen3.5-35B-A3B59.0%
—Agents-A1-4B52.1%
—MiniCPM5-2B43.7%
—LLaDA2.2-mini35.0%