Perceptron Mk1 has documented text, image, and video input; text output; OCR, detection, event-clipping, and embodied-reasoning use cases; a 32K context window; streaming; and OpenAI-compatible chat completions with vision controls. This makes its multimodal I/O stronger than the lower-scoring Gemma 3 anchor’s documented deployment profile, but below Gemini 3.5 Flash, whose 92 reflects substantially broader established multimodal evidence. No independent quality benchmark supports the vendor’s frontier-comparability claim, so technical and reasoning scores remain below Gemini, Kimi K2.5, and Claude Opus 4.8. The listed OpenRouter price of $0.15/$1.50 per million input/output tokens is exceptionally low against published anchors, supporting a high cost-effectiveness score. Pricing clarity is lower because the supplied official materials do not publish an authoritative rate card; the price, 20 tokens/s throughput, and 5.93-second latency come from OpenRouter. The API documentation and familiar OpenAI-compatible interface support solid developer experience, though the closed model has a smaller public integration and operational track record than Gemini or Claude. Coding and agentic scores are deliberately conservative: DeepSWE and LiveCodeBench checked results do not list Mk1, SWE-bench has no direct evidence, and the supplied repository and launch release do not establish Terminal-Bench or Aider performance. Public adoption evidence is similarly thin, with no LMArena or Arena-Hard listing. Documentation substantiates capabilities, not comparative accuracy, reliability, safety, or availability; this creates a materially weaker evidence base than the established anchors.
