Perceptron AI models

Perceptron Mk1.5 is Perceptron's multimodal perception model for image, video, and audio understanding with visual grounding, reasoning, and function calling. It accepts text, images, video, and audio (WAV, MP3, FLAC) and returns text, spatial and temporal annotations (points, boxes, polygons, clips, and video tracks), structured JSON or regex-constrained responses, and function calls for the caller to execute. It supports configurable reasoning effort, multiple assets per conversation, optional video soundtrack analysis, and Multilook for independent prompts over a shared context. The context window is 36,864 tokens.

Perceptron Mk1 ("Mark One") is Perceptron's closed-source vision-language model for video understanding and embodied reasoning. It accepts text, images, and video (MP4, WebM, or pre-decoded frames) and returns text, with optional structured spatial and temporal annotations (points, boxes, polygons, and clips) and optional reasoning. It analyzes video at up to 2 frames per second within a 32K-token context window, and supports grounded Q&A, captioning, OCR, object detection and counting, video clipping, in-context learning from examples, JSON Schema or regex constrained output, and Multilook for up to 16 independent prompts over one shared context. It does not support function calling.