Gemma 4 31B (RadLE 2.0 label)
Google DeepMind · 2026-04-02 · 32.7B parameters
RadLE 2.0 reports an open-cohort Google model only as Gemma 4 31B, without a repository ID, base/instruction-tuned suffix, endpoint, or thinking setting. This benchmark-entry row therefore does not choose between Google's 31B pretrained and instruction-tuned checkpoints and records only metadata shared by both.