FLUX 3 Action enters robot control

// 1h agoMODEL RELEASE

FLUX 3 Action enters robot control

Black Forest Labs released FLUX 3 Action, an open-weight 7B world-action model that jointly predicts robot actions and future video frames from camera input, state, and text instructions. It supports fine-tuning for robots, games, drones, and other interactive environments through LeRobot.

// ANALYSIS

FLUX 3 Action makes BFL’s shift from visual generation to physical intelligence look credible, especially because it targets the real bottleneck in robotics: fast, affordable control.

–Tops RoboLab-120 with 42.92% success, ahead of NVIDIA’s 16B Cosmos 3 Nano at 36.8%
–Uses fewer parameters while retaining joint video-and-action prediction, avoiding the usual WAM latency trade-off
–Public weights, training recipes, and LeRobot integration lower the barrier for independent robotics teams
–Fine-tuning across robot arms, games, and drones suggests a broader visual-control platform beyond industrial manipulation
–Real-world deployment still requires robot-specific action heads, safety limits, simulation testing, and human oversight

// TAGS

flux-3-actionroboticsmultimodalopen-weightsfine-tuningedge-ai

DISCOVERED

1h ago

2026-09-27

PUBLISHED

1h ago

2026-09-27

RELEVANCE

9/ 10

AUTHOR

AI Search

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

LAUNCH53m ago

Tesana Builds Wizarding Game From One Prompt

A Tesana user created a full wizarding-school game featuring classes, spell learning, exploration, quests, and boss fights. The creator credits Three.js and Opus 5.5, showcasing how natural-language game creation is moving beyond simple prototypes.

RESEARCH1h ago

Skild AI trains robots through physical self-play

Skild AI’s S1 robotics foundation model learned soccer in NVIDIA Isaac Sim by competing against earlier versions of itself for more than 140 simulated years. The resulting policy developed dribbling, shielding, tackling, and recovery behaviors before transferring to a humanoid robot.

MODEL3h ago

CLM-8B launches open, fast agent decision model

Stanford and NVIDIA researchers released CLM-8B, an Apache 2.0 contrastive model that scores candidate actions through shared state-action embeddings instead of generating text. Its independently cached action embeddings deliver up to 9× lower latency than Jev on selected agent, gaming, and tool-calling tasks.