Google: Gemma 4 31B

Google's Gemma 4 31B accepts text, image, and video input and returns text, with a context window of 262,144 tokens and a maximum completion length of 16,384 tokens. It supports tool calling and a reasoning mode, which suits agent-style workflows and multi-step tasks. Structured output support is not documented, so teams that depend on strict JSON schemas should test it before committing. At $0.20 per million input tokens and $0.40 per million output tokens, it sits in the low-cost tier, and the long context makes large document or video inputs affordable to process. Its benchmark record is thin: a blended score of 47.9 drawn from only two benchmarks, an AI Index score of 24.2 and an EQ-Bench score of 70.8. The EQ-Bench result points to reasonable conversational handling, but two data points do not establish general capability. Shortlist it for cost-sensitive multimodal or tool-using workloads, and validate quality on your own tasks first.