Mistral: Mistral Small 3.1 24B

Mistral Small 3.1 24B Instruct is a text generation model from Mistral, released by the provider on 17 March 2025. It is an upgraded variant of Mistral Small 3 (2501) and carries 24 billion parameters. On api14 it sits in the text category, and the model id you use when calling it is mistralai/mistral-small-3.1-24b-instruct.

The model accepts text and images as input and returns text. Its context window is 128,000 tokens, with a maximum output of 102,400 tokens, so long documents and long generations both fit comfortably in a single request. The knowledge cutoff is 31 October 2023. Multimodal input means you can pass images alongside text prompts, which is useful when a task depends on information that only exists in a screenshot, diagram or scanned page.

Because it handles both text and images over a large context, the model suits document and image understanding tasks: summarising long reports, extracting structured data from scanned forms or screenshots, answering questions grounded in a supplied document, and mixed text and image support workflows. The 128k context also makes it reasonable for codebase level questions, long conversation histories and batch style analysis where you want to send a lot of material in one call. Tool and function calling support extends this to agent style pipelines that need to fetch data or trigger actions.

You call the model through api14's OpenAI-compatible API using the model id mistralai/mistral-small-3.1-24b-instruct, which means existing OpenAI SDKs and clients work with only the base URL and key changed. The request parameters it supports are frequency_penalty, logit_bias, logprobs, max_tokens, min_p, presence_penalty, repetition_penalty, seed, stop, structured_outputs, temperature, tool_choice, tools, top_k, top_logprobs and top_p. Structured outputs and the sampling controls are useful when you need predictable JSON or tightly constrained generations, while seed helps with reproducibility during testing.

Two limits are worth noting before you commit. The knowledge cutoff is 31 October 2023, so anything after that date has to come from your prompt or from tools rather than from the model's own knowledge. Output is text only, so if you need generated images or audio, this model is not the right fit. For tasks that are purely text and well within a smaller context, a lighter model may be cheaper and faster, but for multimodal input at 128k tokens this one covers a broad range of work.