Black Forest Labs Enters Robotics Race

🔥 What happened Black Forest Labs dropped FLUX 3 Action — a 7B open-weights World Action Model for robot control. It reads camera frames, robot state, and text instructions, then predicts future video frames and the next action chunk together. It ranks first on the RoboLab-120 leaderboard at 42.92% task success. 💡 Why it matters The model beats NVIDIA's Cosmos 3 Nano by 6.1 percentage points — with 56% fewer parameters. In FP8, it runs up to 3.95x faster than Cosmos 3 Nano and plans 2.13 seconds of motion per call, versus 1.0 second for π0.5. Without pretraining, DROID-only training stayed below 1% — with pretraining, the same protocol hit 11.6%. ⚡ Our take BFL just proved World Action Models don't have to be slow. If you're still betting on pure VLAs, it's time to rethink — joint video and action prediction is the new baseline.