The model page says 32,768 natively with performance validated for up to 4x YaRN https://huggingface.co/Qwen/Qwen3-235B-A22B#processing-long-texts https://huggi

The model page says 32,768 natively with performance validated for up to 4x YaRN https://huggingface.co/Qwen/Qwen3-235B-A22B#processing-long-texts https://huggingface.co/Qwen/Qwen3-235B-A22B#processing-long-... That would seem to align with the 131k number?