Perceptron Mk1.5

Perceptron released Mk1.5 on September 25, 2026 as an update to its perception model line aimed at physical/embodied agents. According to the TPS Report write-up, Mk1.5 accepts four input modalities — text, image, video, and audio — and responds with natural-language text plus, when explicitly requested via an annotation format parameter, structured spatial outputs including points, bounding boxes, polygons, and temporal video clips. The model also supports graded reasoning via standard reasoning controls, function tool calling, and structured outputs via JSON Schema, with audio analysis on video soundtracks treated as an opt-in, per-request feature. Per the TPS article and corroborated by aggregator catalog data, Mk1.5 is priced at $0.15 per 1M input tokens and $1.50 per 1M output tokens, with a reported context window of roughly 36K–37K tokens (sources differ between 36K, 36.9K, and 37K). Mk1.5 succeeds Perceptron's prior flagship vision-language model Mk1, retaining its focus on video and embodied reasoning while adding audio input and structured spatial annotations. No third-party benchmark results were available at release.

LM Market Cap's model page for Perceptron Mk1.5 (last updated September 26, 2026) confirms the release date of September 25, 2026 and reiterates Perceptron's positioning of Mk1.5 as an embodied reasoning model for physical agents with text, image, video, and audio inputs and text output augmented by optional structured spatial annotations. The page lists supported capabilities including reasoning, vision, function calling, JSON mode, and streaming. The aggregator reports pricing of $0.15 per 1M input tokens and $1.50 per 1M output tokens, a context window of 36.9K tokens with 8.2K maximum output, and assigns an overall composite score of 40/100, ranking the model 239 of 440 in its "Coding" category. The "Coding" categorization appears to be the aggregator's own framing and is not strongly aligned with the model's stated embodied/perception focus, and like the other sources, this page provides no primary benchmark measurements.

Benchable's catalog listing for Perceptron Mk1.5 confirms the September 25, 2026 release and describes it as Perceptron's embodied reasoning model for physical agents, accepting text, image, video, and audio input and producing text output with optional structured annotations (points, boxes, polygons, tracks). The listing reports a 36K context window, input pricing of $0.15 per 1M tokens and output pricing of $1.5 per 1M tokens, and lists the endpoint slug as perceptron/perceptron-mk1.5-20260925. Benchable indicates that the model supports a standard set of parameters including temperature, frequency/presence penalties, max tokens, top P, tool choice, reasoning controls, and structured outputs. As of the listing, no benchmark results have been recorded against the model ("No results match the current filters"), so performance claims are not yet independently verified; the page also notes the predecessor Perceptron Mk1 (May 12, 2026) is currently listed as unavailable.