model releaseBlack Forest Labs

Black Forest Labs Releases FLUX 3 Action, a 7B-Parameter Open Robotics Model

TL;DR

Black Forest Labs has released FLUX 3 Action, an open-weight robotics model built on its FLUX 3 multimodal foundation. The 7-billion-parameter model reads multi-camera video feeds and predicts what a robot should do next, claiming a record success rate on the RoboLab-120 leaderboard while running nearly 4x faster than the previous best open model.

2 min read
0

Black Forest Labs has released FLUX 3 Action, an open AI model built specifically for robotics control, the company announced. The model is derived from FLUX 3, BFL's multimodal foundation model trained primarily on video data alongside image and audio inputs.

What It Does

FLUX 3 Action functions as a "world-action model": it ingests multi-camera video feeds from a robot's workspace and outputs two things simultaneously — the next action an agent should take, and a prediction of how the environment will change as a result. This combines action planning with world modeling in a single system, rather than relying on separate components for perception, planning, and simulation.

Benchmark Claims

According to BFL, FLUX 3 Action sets a new record for success rate on the RoboLab-120 leaderboard. The company has not disclosed the exact success rate score. What it has specified is the model's size and speed relative to competitors: FLUX 3 Action runs at 7 billion parameters — according to BFL, less than half the size of the previous best-performing open robotics model — while executing up to 3.95 times faster.

BFL frames this efficiency gap as the central design constraint for robotics deployment. Large reasoning models, the company argues, can generate strong plans but are too slow and computationally heavy to run on physical robots that need to react in real time. A 7B model that beats larger systems on success rate while running nearly 4x faster directly targets the latency and on-device compute limits that have kept bulkier models out of production robotics.

Beyond Physical Robots

BFL also points to digital environments as a secondary application. Video games, the company says, serve as testing grounds for navigation and fast-reaction agents — capabilities that could eventually extend to software agents operating computer interfaces rather than physical hardware. No specific product or timeline was given for this direction.

Availability

Model weights for FLUX 3 Action are available now on Hugging Face. Black Forest Labs has not disclosed API pricing or hosted-inference costs; open weights suggest self-hosted deployment is the primary intended use case at launch.

What This Means

FLUX 3 Action is a bet that robotics needs purpose-built, compact models rather than repurposed large reasoning systems. The claimed 3.95x speed advantage at less than half the parameter count of the prior best open model is significant if it holds up under independent testing — RoboLab-120 results and methodology have not yet been verified outside BFL's own reporting. For a company best known for image and video generation (FLUX, FLUX 3), entering robotics signals an attempt to extend its multimodal foundation models into embodied AI, a space currently contested by labs like Google DeepMind (RT-2, Gemini Robotics) and Physical Intelligence. Open-weighting the model lowers the barrier for robotics researchers and startups to test world-action architectures without training from scratch, but real-world validation on physical hardware — not leaderboard numbers alone — will determine whether the efficiency claims translate into deployable systems.

Related Articles

model release

Huawei Cloud releases CloudRobo-WM1.2 world action model, claims first place on RoboTwin 2.0 and LIBERO-Pro 2.0

Huawei Cloud has released CloudRobo-WM1.2, an embodied world action model, and says it will be open source. The company claims first place on the RoboTwin 2.0 and LIBERO-Pro 2.0 robot manipulation benchmarks, with scores of 88.98% and 81.7%.

analysis

Black Forest Labs trains FLUX 3 on images, video and sound together, founders say

Black Forest Labs co-founder Andreas Blattmann and Dustin Podell, who led FLUX Kontext, explain why FLUX 3 trains on images, video and sound together. The Latent Space interview also covers autoregression versus diffusion, video upscaling and Kontext's origins.

model release

Huawei Cloud to unveil CloudRobo-WM1.2, its first Ascend-native world model, on November 16

Huawei Cloud will launch CloudRobo-WM1.2, an embodied world model for robots, at an online event on November 16 (14:00 to 15:30 Beijing time). Its poster bills the model as the first Ascend-native world model from the Industry AI Dream Factory. Parameter counts, benchmarks, availability and pricing have not been disclosed.

model release

Runway releases Praxis-1 robot model with open weights under Apache 2.0, tops World Action Models on RoboLab-120

Runway has released Praxis-1, a robot control model, with open weights and source code under the Apache 2.0 license. It scores 46.2 percent on Nvidia's RoboLab-120 leaderboard, ahead of FLUX 3 Action at 42.9 percent, in tests Runway ran itself.

Comments

Loading...