Is Flora state of the art?Pending benchmarks.Announced Oct 20, 2026; no independent results yet.

The details so far

Lab

NVIDIA

Announced

Oct 20, 2026

API

Not yet

Open weights

Published

Parameters

27B captioner and rewarder, 9B temporal segmenter

Lab

NVIDIA GEAR (Generalist Embodied Agent Research)

Models

Flora-Captioner (Qwen3.5-27B, full fine-tune plus GRPO), Flora-TempSeg (Qwen3.5-9B with LoRA), Flora-Rewarder (Qwen3.5-27B, full fine-tune)

Outputs

Timestamped action instructions, outcome and dense progress rewards with failure reasons, rubric step scores

License

NVIDIA OneWay Noncommercial License; research use only, not for commercial or production use (model cards)

Used for

The data engine for NVIDIA GR00T models (lab statement)

Hugging Face

Repos created Oct 1–2, 2026 (UTC); cards list Oct 15 as the Hugging Face release date; announced Oct 20

Tags marked unsure are not stated on the lab's own pages.

What would let us call it

Verdicts use only independent boards we already track. Two consecutive snapshots are required before any crown.

• Publicly announced with a source
• Usable by the public (API or weights)
• Listed on an independent board we track
• Enough results for a rank in a category
• Holds across two snapshots (crowns only)

What the lab claims

Self-reported. Shown for context; not used in any verdict.

Lab claim: competitive with GPT-6 Astra on internal benchmarks at roughly 125× lower cost, and state of the art on WGO-Bench, RoboReward and Robometer.

Lab claim: labels 200,000 hours of video a day in production on 512 GB300 GPUs.

Sources