← Sinan Lab › Select a model
Filter open embodied models by your constraints
Pick an embodiment, required inputs, licence terms and a GPU. We lay out the candidates by verified fields: compatibility status, parameters, VRAM estimate, evidence completeness. Measurements and sources only, no recommendations; an empty cell means we have not verified it.
Required inputs：
Workbench: how many Hz on this card, at what costPick a GPU, set the target control frequency and hours per day. Achievable Hz is computed from measured p95 (1000 ms ÷ p95); cost per 1k inferences from measured p50 × rental price. Models without a measurement say “untested”; we do not estimate.
Hz h
Pick a GPU first; the workbench columns fill in after that.
40 models match · VRAM estimate = params × 2 bytes × 1.2 (activation/cache headroom); for exclusion only, not a measurement. Measurements come from our runs on rented GPUs (bf16, batch 1, fixed input); each row links back to its raw result file on the model page.
| Model | Compatibility with chosen embodiment | Params | fp16 VRAM est. | Measured p50 / p95 | Achievable Hz | Cost per 1k | Licence | Evidence |
|---|---|---|---|---|---|---|---|---|
| Isaac GR00T N1.7 NVIDIA · 2026-04 | — | 3B | 7 GB | — | — | — | ✓ NVIDIA Open Model License (weights); Apache-2.0 (Isaac-GR00T code) | 8/8 |
| GR00T N1.6 3B NVIDIA · 2026 (unverified) | — | 3.29B | 8 GB | — | — | — | ? Model card tagged NVIDIA License (links to LICENSE file, terms pending verbatim verification) | 5/8 |
| LingBot-VLA 2.0 Robbyant (Ant Group) · 2026-07 | — | 6B | 14 GB | — | — | — | ✓ Apache-2.0 | 8/8 |
| GO-1 (Genie Operator-1) AgiBot · 2025-09 | — | 2.82B | 7 GB | — | — | — | ✗ CC BY-NC-SA 4.0 | 8/8 |
| π0 (pi-zero) Physical Intelligence · 2025-02 | — | 3.3B | 8 GB | — | — | — | ? Apache-2.0 (openpi code); no separate licence statement for weights, repository includes Gemma Terms of Use | 7/8 |
| π0.5 Physical Intelligence · 2025-09 | — | ? | ? | — | — | — | ? Apache-2.0 (openpi code); no separate licence statement for weights, repository includes Gemma Terms of Use | 6/8 |
| SmolVLA Hugging Face (LeRobot) · 2025-06 | — | 450M | 1 GB | — | — | — | ? Weights model card has no licence tag (LeRobot code repository Apache-2.0) | 7/8 |
| OpenVLA Stanford / UC Berkeley / TRI et al. (OpenVLA team) · 2024-06 | — | 7.54B | 18 GB | — | — | — | ? MIT (tagged on code and weights repositories); weights derived from Llama-2, subject to the Llama 2 Community License | 7/8 |
| X-VLA Tsinghua University AIR (2toINF) · 2025-10 | — | 880M | 2 GB | — | — | — | ✓ Apache-2.0 (code and model card) | 8/8 |
| Octo UC Berkeley RAIL et al. · 2024-05 | — | 93M | ? | — | — | — | ✓ MIT (tagged on code and weights repositories) | 8/8 |
| RDT-1B Tsinghua University TSAIL (thu-ml) · 2024-10 | — | 1B | 2 GB | — | — | — | ✓ MIT (code, weights and data all MIT, per README) | 8/8 |
| CogACT Microsoft Research · 2024-12 | — | ? | ? | — | — | — | ✓ MIT (code, weights and data all MIT, per README) | 7/8 |
| SpatialVLA Shanghai AI Laboratory et al. · 2025-01 | — | 4.03B | 10 GB | — | — | — | ✓ MIT (both code README and model card tagged MIT) | 8/8 |
| UniVLA OpenDriveLab (HKU et al.) · 2025-05 | — | 7.54B | 18 GB | — | — | — | ✓ Apache-2.0 (tagged on both code repository and model card) | 8/8 |
| MolmoAct Allen Institute for AI（Ai2） · 2025-08 | — | 8.12B | 19 GB | — | — | — | ✓ Apache-2.0 (code and weights) | 8/8 |
| RynnVLA-001 Alibaba DAMO Academy · 2025-08 | — | 7.01B | 17 GB | — | — | — | ? Apache-2.0 (code and model card); README also states research preview intended for | 7/8 |
| WALL-OSS X Square Robot · 2025-09 | — | 4.22B | 10 GB | — | — | — | ? Apache-2.0 (code repository); weights model card has no licence tag | 7/8 |
| InternVLA-M1 Shanghai AI Laboratory InternRobotics · 2025-09 | — | 3.75B | 9 GB | — | — | — | ✗ Code MIT; weights model card tagged CC BY-NC-SA 4.0 | 8/8 |
| GraspVLA Peking University EPIC Lab / Galbot · 2025-07 | — | ? | ? | — | — | — | ✗ CC BY-NC 4.0 (code badge and model card agree) | 7/8 |
| Being-H0 BeingBeyond (Beijing) · 2025-08 | — | 8.11B | 19 GB | — | — | — | ✓ MIT (code repository and model card) | 8/8 |
| EO-1 IPEC-COMMUNITY (Shanghai AI Laboratory affiliated) · 2025-08 | — | 3.77B | 9 GB | — | — | — | ✓ MIT (model card) | 8/8 |
| LingBot-VLA 4B Robbyant (Ant Group) · 2026-01 | — | 4.2B | 10 GB | — | — | — | ? Apache-2.0 (code repository); weights model card has no licence tag | 7/8 |
| NORA Declare Lab (SUTD) · 2025-04 | — | 3.76B | 9 GB | — | — | — | ? Neither repository nor model card has a licence tag | 7/8 |
| VLA-Adapter OpenHelix team · 2025-09 | — | 500M | 1 GB | — | — | — | ? MIT (code repository); weights model card untagged | 7/8 |
| HPT（Heterogeneous Pre-trained Transformers） MIT (Lirui Wang, Kaiming He et al.) · 2024-09 | — | 227M | 1 GB | — | — | — | ? MIT (code repository); model card has no licence tag | 7/8 |
| LAPA KAIST / University of Washington et al. (Latent Action Pretraining) · 2024-11 | — | 7B | 17 GB | — | — | — | ✓ MIT (tagged on both code and model card) | 8/8 |
| LLaVA-VLA OpenHelix team · 2025-06 | — | 7B | 17 GB | — | — | — | ✓ MIT (code and model card) | 8/8 |
| RoboVLMs（KosMos P.H.） Tsinghua University / ByteDance Research / CASIA et al. · 2024-12 | — | ? | ? | — | — | — | ✓ Apache-2.0 (code and model card) | 7/8 |
| Magma-8B Microsoft Research · 2025-02 | — | 8.9B | 21 GB | — | — | — | ✓ MIT (code and model card) | 8/8 |
| V-JEPA 2-AC Meta FAIR · 2025-06 | — | 1B | 2 GB | — | — | — | ? Mostly MIT, some files Apache-2.0 (per README); HF weights require access request | 7/8 |
| DexVLA Midea AI Research et al. (DexVLA team) · 2025-02 | — | 3B | 7 GB | — | — | — | ✗ Code MIT; ScaleDP pretrained weights model card tagged CC BY-NC 4.0 | 8/8 |
| GR-1 ByteDance Research · 2023-12 | — | ? | ? | — | — | — | ? Apache-2.0 (code repository) | 6/8 |
| Spirit v1.5 Spirit AI · 2026-01 | — | 5.4B | 13 GB | — | — | — | ✓ Code MIT (README); model card tagged apache-2.0 | 8/8 |
| Galaxea G0.5 Galaxea · 2026-06 | — | 2B | 5 GB | — | — | — | ✗ G0.5 Community License (non-commercial + limited patent licence; model card license: other); 2026 | 8/8 |
| RDT2-VQ Tsinghua University TSAIL (thu-ml) · 2025-09 | — | 8.29B | 20 GB | — | — | — | ✓ Apache-2.0 (code and model card) | 8/8 |
| InternVLA-A1.5 Shanghai AI Laboratory InternRobotics · 2026-07 | — | 2.69B | 6 GB | — | — | — | ✗ CC BY-NC-SA 4.0 (code repository and model card agree) | 8/8 |
| VLA-0 NVIDIA Research · 2025-10 | — | 3B | 7 GB | — | — | — | ✗ CC BY-NC 4.0 (code and weights); base model subject to the Qwen Research License | 8/8 |
| MiniVLA Stanford ILIAD · 2024-12 | — | 500M | 1 GB | — | — | — | ? Code MIT; README notes pretrained weights may inherit base-model (Llama Community License) restrictions | 7/8 |
| Genie Envisioner（GE-Act） AgiBot · 2025-08 | — | ? | ? | — | — | — | ? Code split in two: LTX/Cosmos/pipeline/openpi_client parts Apache-2.0, remaining code and da | 6/8 |
| GR00T N1.5 3B NVIDIA · 2025-05 | — | 2.72B | 7 GB | — | — | — | ✗ Model card tagged Nvidia License (page states non-commercial use only) | 8/8 |
Clear direction, grounded judgement.