Gemma 4 31B

Gemma 4 31B is the flagship dense configuration in Google DeepMind's Gemma 4 family, positioned as the largest dense offering within a lineup that also includes E2B and E4B variants aimed at compute- and memory-efficient edge deployment. DeepMind frames the family as their most intelligent open models to date, built from the same research and technology foundation as Gemini 3 with a focus on maximizing intelligence-per-parameter. The 31B variant is shipped as an open-weight release under Apache 2.0 and is identified through the instruction-tuned name gemma-4-31B-it, which Google AI Studio exposes directly from the official Gemma 4 page.

As a multimodal model, Gemma 4 31B accepts both text and image input and produces text output, making it well suited to workflows that combine visual understanding with language generation. It is explicitly designed for advanced reasoning, coding, and agentic tasks, and independent reporting attributes strong benchmark results to the release, including 85.2% on MMLU Pro, 89.2% on AIME 2026 without tools, 80.0% on LiveCodeBench v6, 84.3% on GPQA Diamond, and a Codeforces ELO of 2150. These figures paint a picture of a dense model that competes with much larger systems on knowledge, math, code, and scientific reasoning, while remaining openly licensable for teams that want to self-host or fine-tune a frontier-class model without paying for proprietary weights.