Google: Gemini 3.7 Flash

Google: Gemini 3.7 Flash is a multimodal text model from Google, available on api14 under the model id google/gemini-3.7-flash. It sits in the text category and is positioned by its provider for fast agentic workflows, coding, and complex multi-step reasoning, with an emphasis on responsive performance and reliable multi-step execution. If you are already building against Gemini or you want a Google model behind an OpenAI-compatible interface, this is the one to evaluate.

On the input side, the model accepts text, images, video, files, and audio, so a single request can mix modalities rather than forcing you to pre-process everything into plain text. Output is text only. The context window is 1,048,576 tokens, which is large enough to hold long documents, extended conversation histories, or sizeable code repositories in a single call. The maximum output length is 65,536 tokens, so long generations such as full file rewrites or detailed multi-step plans fit within one response.

Those capabilities point to a fairly specific set of use cases. Multimodal input plus a million-token context suits document and media analysis where you need to reason over screenshots, recordings, or attached files alongside written instructions. The provider's framing around agentic workflows and coding makes it a reasonable fit for tool-driven assistants, code review and generation, and multi-step tasks where the model calls functions and continues from their results. Because it handles audio and video directly, transcription-adjacent and media summarization pipelines can skip a separate ingestion step.

Calling it through api14 works like any other OpenAI-compatible endpoint. Send your request with the model field set to google/gemini-3.7-flash, and the gateway translates it to the provider. The supported request parameters are include_reasoning, max_tokens, reasoning, reasoning_effort, response_format, seed, stop, structured_outputs, temperature, tool_choice, tools, and top_p. In practice that covers tool and function calling through tools and tool_choice, structured JSON output through response_format and structured_outputs, determinism through seed, sampling control through temperature and top_p, and reasoning control through the reasoning, reasoning_effort, and include_reasoning options.

Two things are worth checking before you commit. The output is text only, so if you need generated images or audio you will need a different model. And if your workload is simple classification or short summarization on plain text, the broad modality support and very large context may be more than you need; a smaller, text-only model could be a better match.