Google: Gemma 4 31B (free)

The developer who gets the most from Gemma 4 31B is one prototyping multimodal features who wants capability without a bill during early iteration. It is Google DeepMind's 30.7B dense multimodal model, taking text and image input to produce text, with a 262,144-token context window, a configurable thinking mode, native function calling, and vision, and it is free through the OpenRouter API. A concrete fit is building a proof-of-concept that reads a product photo and a spec sheet, reasons about them with thinking mode enabled, and calls a function to log the result, all before committing to any paid infrastructure. Against Claude, ChatGPT, Gemini, and Claude Sonnet 5, which are polished commercial products at metered or subscription rates, this model differentiates on being an open, no-cost route with a large context window and toggleable reasoning, suited to experimentation rather than guaranteed scale. Ideal for developers and hobbyists validating multimodal ideas who value zero cost and a configurable reasoning switch over service guarantees. The trade-off is that availability and rate limits depend on the upstream provider, so it is a poor match for steady production load. Choose it for prototyping, learning, and bursty personal projects where free access outweighs reliability; move to a paid model once your workload needs dependable capacity and support.