Google: Gemini Flash Latest

Google: Gemini Flash Latest is a text generation model from Google, distributed through api14 under the model id ~google/gemini-flash-latest. As the name and provider description indicate, this entry always redirects to the latest model in the Gemini Flash family, so requests sent to this id resolve to whatever Google currently serves as the newest Flash release rather than to a pinned version. That makes it a convenient choice when you want to track Google's current Flash model without changing your integration each time the family is updated.

The model accepts text, image, video, file and audio as inputs, and produces text as its only output. It has a context window of 1,048,576 tokens and a maximum output of 65,536 tokens, which supports long documents, extended conversations and large multimodal payloads in a single request. Supported request parameters include max_tokens, temperature, top_p, seed, stop, response_format, structured_outputs, tool_choice, tools, reasoning, reasoning_effort and include_reasoning.

Those capabilities point to a fairly wide set of practical uses. Because the input side covers text, images, video, files and audio, you can build assistants that reason over mixed media, such as summarizing a recorded meeting, extracting structured data from scanned documents, or answering questions about a video clip. The large context window suits codebase analysis, long report review and multi-turn chat where earlier turns need to stay in view. Tool calling and structured outputs make it a reasonable fit for agent workflows and for pipelines that need JSON-shaped responses rather than free text. The reasoning controls let you trade off between quick answers and more deliberate step-by-step processing depending on the task.

Calling it through api14 works the same way as any other OpenAI-compatible model: send a chat completions request with the model field set to ~google/gemini-flash-latest. You can pass max_tokens, temperature, top_p, seed, stop, response_format, structured_outputs, tool_choice and tools as usual, plus the reasoning-related parameters reasoning, reasoning_effort and include_reasoning if you want to influence or inspect the model's reasoning behavior. Because the id redirects to the latest Flash model, the exact underlying version may change over time, so if you need strictly reproducible outputs across deployments, a pinned model id may be a better fit. For workloads that fall outside text output, such as image or audio generation, you would need a different model, since this one returns text only.