The First Fully General Computer Action Model
Hacker News
February 23, 2026
AI-Generated Deep Dive Summary
The article introduces FDM-1, a groundbreaking foundation model designed for general computer use. Unlike previous models that relied on vision-language approaches and limited datasets, FDM-1 is trained directly on video data from an extensive 11-million-hour screen recording dataset. This innovative approach allows it to perform complex tasks such as CAD modeling, financial analysis, engineering automation, and even driving a car in real-time at 30 frames per second (FPS). The model’s ability to process long sequences of context—minutes or hours of video—marks a significant leap forward in AI capabilities for computer interaction.
FDM-1 achieves this by using a highly efficient video encoder that compresses nearly two hours of video into just one million tokens, making it 50 times more efficient than previous state-of-the-art models and 100 times better than OpenAI’s encoder. This efficiency enables the model to scale up significantly while maintaining high performance. What
Verticals
techstartups
Originally published on Hacker News on 2/23/2026