Perceptron: Perceptron Mk1

Perceptron Mk1, also called Mark One, is a text model from the provider Perceptron. It sits in the text category and is described by its maker as Perceptron's highest-quality vision-language model for video and embodied reasoning. It was released by the provider on 12 May 2026. If you are evaluating models for a project that mixes language with visual input, this is the model Perceptron positions at the top of its own line.

The model accepts text, image and video as input and returns text only. That combination makes it a vision-language model rather than a pure text generator: you can pair images or video with natural language queries and get detailed visual understanding back. The context window is 32768 tokens, and the maximum output is 8192 tokens. Supported request parameters are frequency_penalty, include_reasoning, max_tokens, presence_penalty, reasoning, structured_outputs, temperature, top_k and top_p. The presence of reasoning and include_reasoning suggests the model can expose a reasoning mode, and structured_outputs lets you constrain the shape of the response.

Those capabilities point to a fairly specific set of use cases. Video and image understanding paired with natural language queries fits tasks such as describing or summarising visual material, answering questions about what happens in a clip, and embodied reasoning, where a model reasons about scenes and physical situations rather than just static text. Because output is text, the model works well as an analysis layer: you feed in visual material plus a question, and you consume a written answer. Structured outputs make it usable in pipelines where the result has to fit a schema, and the reasoning parameters suit problems where you want the model to work through a question before answering.

You call the model through api14's OpenAI-compatible API using the model id "perceptron/perceptron-mk1". Because the interface follows the OpenAI format, you can point an existing OpenAI client at api14 and pass that id in the model field. The request body accepts the parameters listed above: frequency_penalty, include_reasoning, max_tokens, presence_penalty, reasoning, structured_outputs, temperature, top_k and top_p. Use max_tokens to cap the response within the 8192 token output limit, and temperature, top_p and top_k to control sampling.

Two constraints are worth checking before you commit. The context window is 32768 tokens, which is moderate, so very long videos, long transcripts or large document sets may need to be split or summarised before they fit. Output is text only, so the model cannot generate images or video. If your workload is text only with no visual component, a pure text model may be a simpler fit; Mk1 is most useful when images or video are part of the input.