Mistral: Ministral 3 14B 2512

Mistral: Ministral 3 14B 2512 is a text generation model from Mistral, released on 2 December 2025. It is the largest model in the Ministral 3 family, and Mistral describes it as offering frontier capabilities and performance comparable to its larger Mistral Small 3.2 24B counterpart. If you already work with Mistral models, this one sits in the smaller and more efficient part of their lineup while aiming at capability levels usually associated with bigger siblings.

The model accepts text and images as input and returns text. Its context window is 262,144 tokens, and the maximum output is 209,715 tokens, so long inputs and long generations are both practical within a single request. Because it handles image input alongside text, it can work on tasks that mix a prompt with a screenshot, diagram, photo or scanned page, as long as the response you need is written text.

Those capabilities point to a fairly wide set of uses. Long context suits work over large documents, codebases, transcripts or report sets where you want the model to reason across the whole thing rather than a chunk at a time. Image input makes it a fit for document understanding, extracting or summarising information from screenshots and scans, and answering questions grounded in a visual. The large output ceiling helps when the answer itself is long, such as a full draft, a structured extraction over many items, or a detailed analysis. Tool calling and structured outputs make it usable as the reasoning step inside an agent or pipeline that needs to return machine readable results.

You call it through api14's OpenAI-compatible API using the model id "mistralai/ministral-14b-2512". The supported request parameters are frequency_penalty, max_tokens, presence_penalty, response_format, seed, stop, structured_outputs, temperature, tool_choice, tools and top_p. In practice that means you can control sampling with temperature and top_p, discourage or encourage repetition with the penalty parameters, cap the response with max_tokens, stop generation on given sequences, and set a seed for more reproducible runs. For programmatic use, response_format and structured_outputs let you request shaped output, while tools and tool_choice let the model call functions you define.

There is no stated restriction in the facts beyond the modality split: input can include images, but output is text only, so this is not a model for generating images. If your workload is purely text and you want the smallest capable option, a smaller Ministral 3 variant may be a better fit, and if you need image generation rather than image understanding, you will need a different model entirely.