Mistral: Mistral Medium 3.5

Mistral Medium 3.5 is a dense 128B instruction-following model from Mistral AI. It sits in the provider's medium tier, above the smaller models and below the largest ones, and it is built for general text work rather than a single narrow task. The model was released by the provider on 2026-04-30. On api14 it is listed under the model id mistralai/mistral-medium-3-5 and falls into the text category.

The model accepts text, image and file inputs, and it produces text output only. Its context window is 262144 tokens, and the maximum output length is 209715 tokens, so long inputs and long generations are both possible within the same request. Because images and files can be passed in alongside text, the model can work over mixed material such as documents with embedded figures or screenshots paired with instructions. The provider describes it as designed for agentic workflows, coding, and complex tasks, which matches the combination of a large context, multimodal input and a large output budget.

Those capabilities point to a few practical uses. Agentic workflows are a natural fit: the model can hold a long running conversation or task state, call tools, and reason across many steps before producing a final answer. Coding is another, whether that means generating code, explaining an existing file, or working through a repository excerpt supplied as a file input. Complex document tasks also follow from the input modalities and context size, for example summarising a long report, extracting structured data from a scanned page, or answering questions about a set of uploaded files. The reasoning parameters give you a way to trade latency for more deliberate problem solving on harder prompts.

You call the model through api14's OpenAI-compatible API using the model id mistralai/mistral-medium-3-5. Requests follow the familiar chat completions shape, so existing OpenAI client libraries work with only a base URL and key change. The supported request parameters are frequency_penalty, include_reasoning, max_tokens, presence_penalty, reasoning, reasoning_effort, response_format, seed, stop, structured_outputs, temperature, tool_choice, tools and top_p. In practice this covers sampling controls, deterministic runs via seed, stopping conditions, structured or JSON-style output through response_format and structured_outputs, and tool use through tools and tool_choice. The reasoning and include_reasoning parameters, along with reasoning_effort, let you control how much internal reasoning the model performs and whether that reasoning is returned.

There is no stated limitation beyond the output being text only, so if you need image or audio output this model is not the right choice. If your workload is small and latency sensitive, a lighter model in the same family may be a better fit, while Mistral Medium 3.5 suits tasks that need the full context window, multimodal input or tool-driven agent behaviour.