Black Forest Labs Open-Weights FLUX 3 Action, a 7B Robotics Model
TL;DR
- BFL's own technical report scores the single-step checkpoint at 38.3% vs Cosmos 3 Nano's 36.8%; the cited 42.92% headline figure reflects a full multi-step variant, so performance claims vary by checkpoint.
- NVIDIA's public RoboLab-120 leaderboard had not yet listed FLUX 3 Action at press time, blocking independent third-party score verification at the moment of announcement.
- The quantized FP8 variant runs on 24 GB consumer GPUs, putting competitive robot policy development within reach of teams without H200 access.
Black Forest Labs has open-weighted FLUX 3 Action, a 7-billion-parameter world-action model, and reports it scores 42.92% on NVIDIA's RoboLab-120 benchmark, 6.1 points above the previous leader, NVIDIA's own 16-billion-parameter Cosmos3-Nano-Policy. Per VentureBeat, the model runs 1.43 times faster than Cosmos while using 44% of its parameters. The prior top score of 39.0% belonged to OASIS WAM.
The benchmark is NVIDIA's, which is what makes the result awkward. VentureBeat describes RoboLab-120 as "an NVIDIA-developed simulation benchmark intended to test general-purpose robotic policies across 120 tasks, spanning visual understanding, relational reasoning and procedural skills."
FLUX 3 Action ingests camera frames, robot state and text instructions, and predicts the next 32 actions plus how the scene will evolve visually. BFL says it can be fine-tuned from roughly 200 teleoperated episodes. The release ships weights, code, a fine-tuning recipe, benchmarks, reproducible examples and integration with Hugging Face's LeRobot framework. Missing from the release: an Action API, and a specific commercial license for the weights.
BFL has prior industrial ground via FLUX-mimic, built with Swiss startup mimic robotics, which the article reports was tested in Audi production environments for industrial manipulation tasks. Drone-flight and Doom-gameplay demonstrations are characterized as "early experiments rather than production capabilities."
BFL closed a $300 million Series B at a $3.25 billion valuation in December 2025, taking total funding above $450 million. The release lands in a busy open-robotics week that also included Alphabet's Intrinsic open-sourcing its Intrinsic Core stack at ROSCon.
What others are reporting
-
Black Forest Labs Read →
First-party technical report with ablation studies, a public fine-tuning recipe, and hybrid cost analysis showing 53% per-dollar savings when FLUX 3 Action is paired with a reasoning model.
Our single step 7B checkpoint outperforms every other open policy on RoboLab at 38.3%, compared to 36.8% for Cosmos 3 Nano, with 1.34x through 2.28x speed improvement.
-
Hugging Face Read →
Model card specifying exact hardware minimums, tensor formats, sampling parameters, six checkpoint variants, and an explicit safety constraint requiring human oversight before any deployment.
Runs in about 32 GB of GPU memory in bfloat16 on an NVIDIA H200; with fp8 quantization and encoder offload it fits 24 GB cards.
-
AlphaSignal Read →
Breaks down WAM vs VLA architectural tradeoffs and includes an independent Franka arm evaluation by Positronic Robotics, adding third-party hardware validation absent from the announcement.
Runs 1.45x to 3.95x faster than competing WAMs by predicting a 2.13s action horizon jointly with video.
-
Crypto Briefing Read →
Flags that no 7B model was visibly listed among top contenders on the public leaderboard at publication, introducing the first published skepticism of BFL's top-ranking claim.
The model reportedly outperforms larger competitors on the RoboLab-120 benchmark, though there is no public confirmation yet.
-
Superpower Daily Read →
Notes that promised weights and license documentation were inaccessible after the announcement, making independent reproduction impossible and corroborating the leaderboard gap.
NVIDIA's public leaderboard had not yet added FLUX 3 Action when VentureBeat checked before publication.
Originally reported by venturebeat.com
Read the original article →Original headline: Black Forest Labs Open-Weights FLUX 3 Action, a 7B World-Action Model Topping Nvidia RoboLab-120