presentofai

Black Forest Labs launches FLUX 3 Action open-weights robotics model

TL;DR

Black Forest Labs released FLUX 3 Action, an open-weights robotics action model that leads its category leaderboard at half the parameter count of rivals, with live demos controlling drones and completing Doom without dying.

What happened

  • Black Forest Labs debuted FLUX 3 Action on September 23, 2026, targeting the robotics action-model segment.
  • The model is open-weights, making it freely inspectable and deployable outside a proprietary API.
  • FLUX 3 Action tops its category leaderboard despite running at half the parameter count of competing models.
  • Task-specific policies derived from the model were demonstrated on two real-world tasks: flying physical drones and completing the game Doom with zero deaths.

Why it matters

  • Efficiency at the frontier is the headline: matching or beating larger models at half the size compresses the compute cost of deploying robotics AI at scale.
  • Open weights shifts leverage to hardware makers, robotics startups, and researchers who can fine-tune without licensing fees or API dependency.
  • The drone demo signals readiness for real physical-world control loops, not just simulation benchmarks, raising the credibility bar for the release.
  • Black Forest Labs, already known for image generation with FLUX, is now staking a claim in embodied AI, broadening competitive pressure on robotics-focused labs and foundation-model incumbents.
  • A leaderboard-topping open model creates a new baseline that proprietary competitors must now publicly beat or explain away.

What to watch next

  • Whether robotics hardware partners (drone manufacturers, industrial arm vendors) announce integrations or fine-tuned variants built on FLUX 3 Action weights.
  • How competing labs (those whose larger models were just outperformed) respond: expect benchmark rebuttals or accelerated releases.
  • Whether Black Forest Labs publishes training data and methodology, which would determine how reproducible the efficiency gains are and how quickly the community can extend them.

Originally published on Present of AI, a daily source-linked AI news timeline. Read the full timeline or browse the open dataset.