Breaking: Perceptron Mk1 Launches at Fraction of Competitor Pricing

A two-year-old startup, Perceptron Inc., today released its flagship video analysis AI model, Mk1, at a cost 80-90% lower than offerings from Anthropic, OpenAI, and Google. The model, available via API at $0.15 per million input tokens and $1.50 per million output tokens, aims to democratize real-time video understanding for enterprises.

Unprecedented Price-Performance Ratio

“We built Mk1 from the ground up to handle the complexities of the physical world without the exorbitant costs tied to competitors,” said Armen Aghajanyan, CEO and co-founder of Perceptron, formerly of Meta FAIR and Microsoft. “This model can identify inconsistencies in marketing videos, flag security threats, and even analyze body language — at a price that finally makes sense for large-scale deployment.”

Industry analysts note the price gap is striking. Anthropic’s Claude Sonnet 4.5, OpenAI’s GPT-5, and Google’s Gemini 3.1 Pro all cost significantly more per token, often exceeding $1.50 per million input tokens.

Benchmark Dominance in Spatial and Video Reasoning

Mk1’s performance matches or surpasses leading models on multiple industry benchmarks. On spatial reasoning tasks, it scored 85.1 on EmbSpatialBench, beating Google’s Robotics-ER 1.5 (78.4) and Alibaba’s Q3.5-27B (84.5). In the specialized RefSpatialBench, Mk1’s score of 72.4 dwarfs GPT-5m’s 9.0 and Sonnet 4.5’s 2.2 — a massive leap in referring expression comprehension.

Video benchmarks tell a similar story. On the EgoSchema “Hard Subset,” Mk1 matched Alibaba’s Q3.5-27B with a score of 41.4, while Google’s Gemini 3.1 Flash-Lite managed only 25.0. On the VSI-Bench, Mk1 achieved 88.5 — the highest score recorded among compared models, validating its ability to handle temporal reasoning.

Designed for Real-World Applications

Beyond benchmarks, Perceptron emphasizes practical use cases: live security monitoring, automatic clipping of marketing video highlights, flagging video gaffes, and analyzing participant behavior in controlled studies. A public demo site is available for testing, and enterprise customers can begin integrating Mk1 immediately.

Background

Perceptron spent 16 months developing Mk1, focusing on a proprietary “multi-modal recipe” that addresses cause-and-effect, object dynamics, and the laws of physics. The startup, led by former Meta and Microsoft researcher Aghajanyan, saw a gap in the market: existing video AI models were either too costly or lacked robust spatial-temporal understanding. Mk1 was built specifically to excel on the “Efficiency Frontier,” balancing high performance with low cost.

What This Means

This launch signals a shift in the AI industry toward cost-efficient reasoning models that understand the physical world. As enterprises seek to deploy video analysis at scale — from security to marketing to research — Perceptron’s pricing could force competitors to reassess their token costs. “We expect Mk1 to set a new baseline,” Aghajanyan said. “The message is clear: high-performance AI no longer requires a premium price tag.”

Interested users and potential enterprise customers can try it out for themselves on Perceptron’s public demo site here.