Big techChina
ByteDance's research division (formerly ByteDance Research for robotics) that develops the GR series of video-pretrained and vision-language-action robot manipulation models.
ByteDance Seed is ByteDance's AI research division, whose robotics group develops generalist robot models. In July 2025 it released GR-3, a vision-language-action model, together with ByteMini, a bi-manual mobile robot that runs it. In 2026 the Seed Robotics team was consolidated under a multimodal interaction and world-modeling group led by Zhou Chang.
Sources: seed.bytedance.com (22 Jul 2025) · theinsight.asia (25 Aug 2026)
Corporate division · AI research division of ByteDance; no separate funding
#28 of 104 model companies · score 57.7 / 100 · updated 19 Oct 2026
Each bar shows an input from 0 to 100% before weighting. 1 of 5 listed models have open weights. Most-starred official repository: bytedance/GR-1 (308 stars, 19 Oct 2026).
The score (0–100) measures public, checkable evidence about a company's robot models. It does not measure private capability. A company that publishes no benchmark results and no weights can still have strong models. Every input is computed from the data on each company page, and every number there links to its source.
| Input | Weight | How it is measured |
|---|---|---|
| Published benchmarks | 35% | For each scored benchmark, the company's best result divided by the best result any company has published, averaged over the benchmarks it reports. No published results scores 0. Results come only from the model's own paper, model card or the official benchmark leaderboard. Scored: LIBERO (4-suite average success, %); SimplerEnv Google Robot, visual matching (%); SimplerEnv Google Robot, variant aggregation (%); SimplerEnv WidowX / Bridge (%); CALVIN ABC→D (average length, 0–5); RoboArena (pairwise rating). |
| Open weights | 20% | Share of the company's listed models whose weights the public can download. |
| Adoption | 20% | GitHub stars on the company's most-starred official model repository, on a log scale relative to the most-starred repository in the list. Hugging Face downloads are shown on company pages but not scored, because gated repositories report zero. |
| Released models | 15% | Number of publicly announced, named models, capped at 5. |
| Disclosed funding | 10% | The sum of the funding rounds listed on the company's profile that were disclosed in USD, on a log scale from $10M to the largest sum in the list. Rounds reported only in other currencies are listed but not converted or counted. This input applies only to private startups and robot makers. For big-tech divisions, research labs and listed companies it is n/a, and its weight is shared across the other inputs in proportion. |
Sources. arXiv papers and official model cards for benchmarks and parameters; official leaderboards (RoboArena); the GitHub and Hugging Face APIs for stars and downloads; company announcements and established press for funding. Missing values are shown as n/a, never estimated.
Updated 19 Oct 2026. Benchmark protocols differ between papers, for example in fine-tuning and number of trials, so small gaps between close scores are not meaningful.
Open weightsPolicy
GPT-style manipulation policy pretrained on large-scale video generation and fine-tuned on robot data; CALVIN checkpoints published in the code repo.
| Benchmark | Result | Source |
|---|---|---|
| CALVIN ABC→DAverage successful length (0-5) | 3.06 | arxiv.orgTable 1 (ABCD->D reported as 4.21 in the same table). As of 21 Dec 2023 |
Closed weightsVLA
Multi-stage pipeline (demonstration filtering, symmetry augmentation, online RL) that specializes a generalist VLA for long-horizon dexterous tasks; reports 83.3% success on real-robot shoe lacing.
No benchmark results on a shared public benchmark have been published for this model.
Closed weightsVLA
Vision-language-action model co-trained on web vision-language data, VR human trajectories and robot data, deployed on the ByteMini bimanual mobile robot; evaluated on real-world tasks only.
No benchmark results on a shared public benchmark have been published for this model.
Closed weightsVLA
Generative video-language-action model pretrained on about 38 million video clips and fine-tuned on robot trajectories.
| Benchmark | Result | Source |
|---|---|---|
| OtherCALVIN ABCD->D average successful length (0-5)Shown, not scored | 4.64 | arxiv.orgFigure 10; paper reports the ABCD->D split only, not ABC->D. As of 8 Oct 2024 |
Weights: n/aVLA
Robot imitation policy built on OpenFlamingo vision-language models with an explicit policy head, from ByteDance Research with Tsinghua, SJTU and NUS.
| Benchmark | Result | Source |
|---|---|---|
| CALVIN ABC→DAverage successful length (0-5) | 2.48 | arxiv.orgTable 1, ABC (Lang) -> D. As of 5 Feb 2024 |
