Big techLondon, United KingdomFounded 2010
Alphabet's AI lab; builds the RT series and the Gemini Robotics family of VLA, on-device and embodied-reasoning models.
Google DeepMind is Alphabet's AI research unit, formed by combining DeepMind (started in 2010) with Google Brain, and led by co-founder and CEO Demis Hassabis. DeepMind was a London-based company acquired by Google in January 2014. Its robotics work includes RT-2, a vision-language-action model, and the Gemini Robotics model family; its robotics team is headed by Carolina Parada.
Sources: deepmind.google (17 Oct 2026) · techcrunch.com (26 Jan 2014) · blog.google (1 Apr 2025)
Corporate division · Research division of Google (Alphabet); DeepMind was acquired by Google in January 2014. No separate funding.
#49 of 104 model companies · score 35.4 / 100 · updated 19 Oct 2026
Each bar shows an input from 0 to 100% before weighting. 2 of 13 listed models have open weights. Most-starred official repository: google-deepmind/open_x_embodiment (2.1k 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
RT-1 trained on the cross-embodiment Open X-Embodiment dataset; TensorFlow and JAX checkpoints released.
No benchmark results on a shared public benchmark have been published for this model.
Open weightsPolicy
Robotics Transformer trained on ~130k real-robot episodes covering 700+ tasks; code and checkpoints published (repository archived April 2025). Released by Google Research / Robotics at Google before the DeepMind merger.
No benchmark results on a shared public benchmark have been published for this model.
Closed weightsVLA
VLA for whole-body humanoid control (demonstrated on Apptronik Apollo 2 with 22-DoF hands) and bi-arm Franka Duo from a single checkpoint; available to early-access partners.
No benchmark results on a shared public benchmark have been published for this model.
Closed weightsEmbodied reasoning
Embodied-reasoning model based on Gemini 3.5 Flash for spatial reasoning, multi-step planning and multi-robot coordination; preview in Google AI Studio and the Gemini API.
No benchmark results on a shared public benchmark have been published for this model.
Closed weightsVLA
Lightweight on-device VLA based on Gemma models and Gemini Robotics 1.5 technology, evaluated on SO101, Dexmate and Trossen bi-arm platforms; trusted testers only.
No benchmark results on a shared public benchmark have been published for this model.
Closed weightsEmbodied reasoning
Updated embodied-reasoning preview model in the Gemini API adding instrument reading and improved spatial and physical reasoning; scheduled for shutdown 2026-08-31 in favour of ER 2.
No benchmark results on a shared public benchmark have been published for this model.
Closed weightsVLA
VLA that reasons before acting and transfers motions across ALOHA 2, bi-arm Franka and Apptronik Apollo; available to select partners.
No benchmark results on a shared public benchmark have been published for this model.
Closed weightsEmbodied reasoning
Embodied-reasoning VLM for spatial understanding and multi-step task planning, available to developers in preview via the Gemini API (preview shut down 2026-04-30).
No benchmark results on a shared public benchmark have been published for this model.
Closed weightsVLA
VLA optimised to run locally on robot hardware, trained on ALOHA and adapted to bi-arm Franka FR3 and Apptronik Apollo; released to trusted testers with a Gemini Robotics SDK.
No benchmark results on a shared public benchmark have been published for this model.
Closed weightsVLA
Gemini 2.0-based VLA trained mainly on the bi-arm ALOHA 2 platform, also shown on bi-arm Franka and adapted to the Apptronik Apollo humanoid; offered to partners only.
No benchmark results on a shared public benchmark have been published for this model.
Closed weightsEmbodied reasoning
Gemini 2.0-based vision-language model specialised for spatial and embodied reasoning (pointing, 3D detection, grasp and trajectory prediction); initially offered to trusted testers.
No benchmark results on a shared public benchmark have been published for this model.
Closed weightsVLA
RT-2 trained on the Open X-Embodiment dataset; weights not released (only RT-1-X checkpoints are in the public repo).
No benchmark results on a shared public benchmark have been published for this model.
Closed weightsVLA
Vision-language-action model that co-fine-tunes a web-scale VLM (PaLI-X or PaLM-E) to output robot actions as text tokens.
No benchmark results on a shared public benchmark have been published for this model.
