{"id":"fineart-vla","slug":"fineart-vla","name":"FineART-VLA","orgId":"scale-ai","family":"fineart-vla","releaseDate":null,"availability":"announced","openWeights":true,"license":null,"paramsTotal":null,"paramsSource":null,"contextWindow":null,"knowledgeCutoff":null,"modelCardUrl":"https://labs.scale.com/blog/fineart","hfRepo":null,"modalities":["robotics"],"aliases":["FineART-VLA","fineart vla","fineart-vla","lerobot/FineART-VLA-subtask-midtrained","FineART-VLA-subtask-midtrained","fineart-vla-subtask-midtrained","scale fineart-vla"],"aliasNote":null,"origin":"tracked","tracked":{"trackedSince":"2026-10-29","announced":"2026-10-29","blurb":"An open vision-language-action (VLA) robot model from Scale AI and Hugging Face's LeRobot team that plans before it acts: a System 2 part predicts the next subtask in words and a System 1 part turns it into arm actions. Fine-tuned from Physical Intelligence's pi0.5 on FineART, Scale's dataset of 40,543 two-arm robot episodes with 533,913 subtask labels. About 4.1B parameters; weights on Hugging Face under the Gemma license.","api":"none","openWeights":"yes","params":{"text":"About 4.1B (4,143,404,816 in the safetensors): a Gemma 2B vision-language backbone plus a Gemma 300M action expert","source":"Hugging Face model card and safetensors metadata"},"facts":[{"label":"Lab","value":"Scale AI (Scale Labs) with Hugging Face's LeRobot team; published in the lerobot Hugging Face org"},{"label":"Type","value":"Vision-language-action robot policy, not a chat model: camera frames, a language instruction and robot state in, actions out"},{"label":"Design","value":"System 2 predicts subtasks in natural language; System 1 turns the current subtask into actions (50-step action chunks, 10 flow-matching steps)"},{"label":"Base model","value":"lerobot/pi05_base, LeRobot's port of Physical Intelligence's pi0.5 (PaliGemma-based)"},{"label":"Robot","value":"Stationary two-arm ALOHA-style setup: three cameras (high, left wrist, right wrist) at 224×224, 14-dimensional state and action"},{"label":"Training data","value":"FineART dataset (CC-BY-4.0): 40,543 episodes, 1,718 hours, 151 tasks, 533,913 subtask annotations (about 13 per episode)"},{"label":"Weights","value":"Hugging Face, about 9.4 GB (F32 and BF16 safetensors) under the Gemma license inherited from PaliGemma/pi0.5; LeRobot code is Apache-2.0"},{"label":"Code","value":"Training and fine-tuning through LeRobot (lerobot-train; policy fineart_vla); lerobot-annotate labels subtasks on other LeRobot datasets"},{"label":"Paper","value":"arXiv 2609.36416, first posted Sep 29, 2026 (UTC); the weights followed on Oct 29, 2026"}],"claims":["Lab claim: FineART is the largest open robot dataset with subtask labels.","Lab claim: across 3,400 real-world evaluations, subtask training lifts spatial-instruction success from 32.0% to 100.0% and, with step-by-step guidance, unseen long-horizon task success from 16.0% to 76.0%.","Lab claim: on a new robot (YAM arms) it matches the baseline with 10x less fine-tuning data."],"sources":[{"kind":"lab","title":"Scale Labs blog: FineART, Scale AI's first open-source robotics dataset and model","url":"https://labs.scale.com/blog/fineart","date":"2026-10-29"},{"kind":"lab","title":"Scale Labs announcement on X","url":"https://x.com/ScaleAILabs/status/2115837725719048535","date":"2026-10-29"},{"kind":"repo","title":"lerobot/FineART-VLA-subtask-midtrained on Hugging Face","url":"https://huggingface.co/lerobot/FineART-VLA-subtask-midtrained"},{"kind":"repo","title":"lerobot/FineART dataset on Hugging Face","url":"https://huggingface.co/datasets/lerobot/FineART"},{"kind":"lab","title":"LeRobot docs: FineART-VLA","url":"https://huggingface.co/docs/lerobot/main/fineart_vla"},{"kind":"paper","title":"FineART paper (arXiv 2609.36416)","url":"https://arxiv.org/abs/2609.36416"}],"firstBenchmark":null},"organization":{"id":"scale-ai","name":"Scale AI","slug":"scale-ai","country":null,"url":"https://scale.com","color":"#11110F","mark":"S"},"price":null,"answer":{"word":"Pending benchmarks.","scope":"Announced Oct 29, 2026; no independent results yet.","lead":"As of Nov 3, 2026, FineART-VLA is pending benchmarks on isitsota: announced Oct 29, 2026, with no independent results on the boards we track yet.","confidence":"low"},"status":"PENDING_BENCHMARKS","verdicts":[]}