Two-year-old startup Perceptron Inc. released Mk1, a video analysis reasoning model priced at $0.15 per million input tokens — 80-90% cheaper than Anthropic’s Claude Sonnet 4.5, OpenAI’s GPT-5, and Google’s Gemini 3.1 Pro. Despite the low cost, Mk1 matches or exceeds these frontier models on spatial reasoning and video benchmarks, including a top score of 88.5 on VSI-Bench.
What Is Perceptron Mk1?
Mk1 (Mark One) is a proprietary video analysis reasoning model that processes native video at up to 2 frames per second across a 32K token context window. Its key differentiator is “Physical Reasoning” — high-precision spatial awareness that allows the model to understand object dynamics and physical interactions in real-world settings.
How Does Mk1’s Performance Compare to Frontier Models?
Benchmark results show Mk1 occupies a unique position:
- Spatial Reasoning (ER Benchmarks): 85.1 on EmbSpatialBench, surpassing Google’s Robotics-ER 1.5 (78.4)
- RefSpatialBench: 72.4 — a massive leap over GPT-5m (9.0) and Sonnet 4.5 (2.2)
- EgoSchema Hard Subset: 41.4, matching Alibaba’s Q3.5-27B
- VSI-Bench: 88.5 — the highest recorded score among all compared models
What Real-World Applications Does Mk1 Enable?
Early adopters are using Mk1 for diverse applications including auto-clipping highlights from live sports, multimodal quality control agents on manufacturing lines, and wearable assistants on smart glasses that provide context-aware help.
Key Takeaways
- Mk1 priced at $0.15/M input tokens — 80-90% cheaper than frontier rivals
- Matches or exceeds GPT-5, Sonnet 4.5, and Gemini 3.1 Pro on key benchmarks
- Processes native video at up to 2 FPS with 32K context window
- Physical reasoning capability for spatial and object dynamics understanding
- Perceptron also maintains open-weights “Isaac” model series
- Dual-track strategy: closed-source Mk1 API + open-weights Isaac
Frequently Asked Questions
How can Mk1 be so much cheaper than competitors? Perceptron has not disclosed full architectural details, but the efficiency gains appear to come from model architecture optimizations rather than sacrificing capability.
Is Mk1 available as open source? No, Mk1 is a closed-source model accessed via API. However, Perceptron also maintains the Isaac series as an open-weights alternative for edge deployments.
What hardware does Mk1 require? Mk1 is cloud-based via API. The Isaac series is optimized for edge and low-latency deployments, including a 2-billion parameter vision-language model.