Model Releases · The AI Post

Model Releases · Figure

Figure reports 56 percent success for humanoid in 30 unseen homes

Lollipop chart. Pretraining lifts full-task completion from 9 to 56 percent. By task, bed-making reaches 67 percent, folding towels 62, tidying toys 40. — Figure reports full completion only, with no partial credit and every safety intervention counted as a failure.

The company rented the Bay Area properties for the test, and says an identical model trained without its human-video dataset managed 9 percent.

8 sources: Figure — Helix 2.5 announcement, @Figure_robot, @adcock_brett, @coreylynch, Humanoids Daily, @humanoidsdaily, Unite.AI, @SawyerMerritt

Model Releases · Amazon

Moonshot puts Kimi K3 on Amazon Bedrock with a million‑token window

Moonshot describes it as the first open model at 2.8 trillion parameters. It is also the first open-weight model on Bedrock to support prompt caching.

3 sources: AWS, AWS Machine Learning Blog, Unite.AI