Gemma 4 31B IT

Gemma 4 31B IT is the largest model in Google DeepMind's Gemma 4 family of open models, a line the company describes as built from Gemini 3 research and technology to maximize intelligence per parameter. The 31B IT variant is characterized as a dense, instruction-tuned vision-language model, reflecting Gemma 4's lineage of bringing frontier research into openly released weights. It sits alongside smaller efficiency-oriented variants in the same family, but is positioned as the flagship for users who want the strongest reasoning and multimodal understanding the open release offers.

In independent reporting, Gemma 4 31B IT is presented as a practical fit for coding assistance, agentic workflows, document extraction, and question answering, pairing vision-language inputs with text generation. Reported benchmark figures for the model include 2,150 Codeforces ELO, 89.2% on AIME 2026, 84.3% on GPQA Diamond, and 76.9% on MMMU Pro, indicating strong performance across competitive programming, mathematics, graduate-level reasoning, and multimodal understanding. This combination of open availability, dense 31B-parameter design, and instruction tuning makes it well suited to teams building assistive coding agents or document-aware applications that need a balance of capability and openness.