Skip to main content
Perceptron Mk1.5 accepts text, images, video, and audio. It generates text, spatial and temporal annotations, structured responses, and function calls for your application to execute.

​
Specifications

Audio shares the context window with the rest of the input and the output budget. The audio limit is specific to the current Mk1.5 model; a request can reach the context limit before reaching the per-item audio limit. Video soundtracks require vision_config.enable_audio_in_video: true. Responses contain text and structured content; the model does not generate audio. See Audio for input forms and limits.

​
Pricing

Multilook reuses a shared context within one request. Cached tokens are a subset of prompt tokens; do not add them to the prompt-token total a second time. Use the tokenization guide to budget media, reasoning, and conversation history, and the scaling guide to manage throughput.

​
Choose the output for your task

track is a container in generated markup, not an annotation_format value. The optional asset_idx attribute identifies a media occurrence among the assets available at that point in the conversation; it is not a persistent file identifier. If no explicit or inherited selector is present, it defaults to the last asset available when the annotation was produced. See the asset resolution rules.

​
Integration notes

  • Use n: 1 for chat completions. Multilook has separate sampling controls.
  • Inspect finish_reason before using an answer. length means the output may be incomplete; tool calls require a caller-managed round trip.
  • Use reasoning_effort for new reasoning integrations.
  • Function tools cannot be declared in the same request as JSON Schema or regex constrained output. Request a constrained final answer after the tool loop.