Google: Gemma 4 31B

Gemma 4 31B is a multimodal model from Google that accepts image, video, and text inputs. It supports tool use and reasoning, and its context window extends to 262,144 tokens, making it suitable for long-document workflows. Structured output support is unconfirmed from available data, so treat that capability as unverified before committing to a pipeline that requires it. At $0.12 per million input tokens and $0.35 per million output tokens, Gemma 4 31B sits at the budget end of the market, which is its clearest advantage. Its blended benchmark score of 56.4 across only four benchmarks is a limited signal; coding scores are relatively strong at 71.7, but agentic performance is low at 23.8, making it a poor fit for autonomous task pipelines. Teams doing cost-sensitive coding assistance or long-context document processing have reasonable grounds to shortlist it, but buyers needing broad, validated performance should note the thin benchmark coverage.