Latest AI news · Original source attached
FLUX 3 Action is an open world action model
Black Forest Labs releases a 7B open-weight model that turns camera frames and text instructions into the next two seconds of actions.
Reported from Black Forest Labs
Summary written from the original first-party source — the link to it is at the end of this post.
Black Forest Labs released FLUX 3 Action, an open-weights world action model that takes a camera frame and a text instruction and returns the next two seconds of actions. Fine-tuned on the DROID dataset, it places first on the RoboLab benchmark at 42.92% success.
What’s actually new
- Open weights. The 7-billion-parameter model ships under the FLUX Kommunity License v1.0, with code, fine-tune recipes, and both DROID and SO-101 checkpoints integrated into LeRobot.
- Top of the leaderboard. It beats NVIDIA’s 16B Cosmos 3 Nano policy (36.8%) and closed variants like OASIS (39.0%) on RoboLab despite being a fraction of their size.
- Easy to adapt. The same weights were fine-tuned on a few hundred recorded episodes to control robot arms, a Quake-style shooter, a racing game, and an indoor drone, including rooms and sentence phrasings it had never seen.
Why it matters
FLUX 3 Action shows how open world models could standardize a look, understand, act loop for embodied AI. Because it is fully open and fine-tunable, labs and hobbyists can build robot or game policies without closed APIs. It also signals that the open ecosystem is now competitive at the very top of the RoboLab leaderboard.
(Original source: Black Forest Labs)
AI Slop Weekly
Get this story in context, every Sunday.
The monthly rundown of what the stories above add up to — what shipped, what changed, what to actually use. 3-minute read, free forever.
Subscribe freeNo spam. Unsubscribe anytime.