Dream Labs is targeting a broader user base by supporting affordable hardware, such as the $400 Microduck robot from Hugging Face, to move past the industry's current limit of 30,000 existing users. The API is currently available for $3 per hour and enables the control of bimanual robots on tasks for which they had no prior training. Jang, who previously led world models at Nvidia, stated that the company plans to add capabilities for mobile manipulation and dexterous hands as hardware costs decline.
Tuesday, October 20, 2026
Story
Dream Labs Dream-1 Becomes First Robot Model to Top 50% on RoboDojo
AI21 hours ago
A new fundraise of $90M will support a foundation system that can manage two armed machines in entirely new environments. The software, called Dream-1, has reached a record 59% on the RoboDojo-RC Tier 1 benchmark, surpassing the previous 48% held by Google's Gemini 4 Argon. CEO Yoel Jang reported that 70,000 inference calls came from users in 22 countries within five hours of the API launch. The model is 1.7x faster and scores 1.3x higher than GPT-6 Astra Ultrafast, while operating 16 to 40 times cheaper than LLMs from the largest AI laboratories.
Looking for more context or analysis?
Tilt chat provides reliable, tailored insights, alerts, and automations.