Robot Manipulation on LIBERO LONG
97.4Success RateAnchorRefine (X-VLA)
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| AnchorRefine (X-VLA)Size=1.1B2026.04 | 97.4 | — | — | — | — | — | — | — | |
| π0.5 with TAPTBase Model=π0.5, Adaptation=TAPT, Deployment mode=VLA tool-family interface2026.05 | 97.2 | — | — | — | — | — | — | — | |
| X-VLA*Size=0.9B2026.04 | 95.8 | — | — | — | — | — | — | — | |
| OpenVLA-OFT with TAPTBase Model=OpenVLA-OFT, Adaptation=TAPT, Deployment mode=VLA tool-family interface2026.05 | 95.6 | — | — | — | — | — | — | — | |
| CortexParadigm=Agentic, Zero-shot=true, Shared System-1 executor=π0.52026.07 | 95.5 | — | — | — | — | — | — | — | |
| AtomicVLASize=–2026.04 | 95.2 | — | — | — | — | — | — | — | |
| OpenVLA-OFTParadigm=End-to-End, Zero-shot=true2026.07 | 94.5 | — | — | — | — | — | — | — | |
| MemoryVLAParadigm=End-to-End, Zero-shot=true2026.07 | 93.4 | — | — | — | — | — | — | — | |
| π0.5Base Model=π0.5, Adaptation=SFT, Deployment mode=Standalone2026.05 | 92.4 | — | — | — | — | — | — | — | |
| π0.5Paradigm=End-to-End, Zero-shot=true2026.07 | 92.4 | — | — | — | — | — | — | — | |
| UniVLASize=9B2026.04 | 92 | — | — | — | — | — | — | — | |
| OpenVLA-OFTBase Model=OpenVLA-OFT, Adaptation=SFT, Deployment mode=Standalone2026.05 | 92 | — | — | — | — | — | — | — | |
| Gemini-3.1-ProParadigm=Agentic, Zero-shot=true, Shared System-1 executor=π0.52026.07 | 91 | — | — | — | — | — | — | — | |
| SeerSize=0.3B2026.04 | 87.7 | — | — | — | — | — | — | — | |
| π0Size=3B2026.04 | 85.2 | — | — | — | — | — | — | — | |
| BlockVLAType=Block Diffusion, Training Steps=50k2026.05 | 85.2 | — | — | — | — | — | — | — | |
| π0Paradigm=End-to-End, Zero-shot=true2026.07 | 85.2 | — | — | — | — | — | — | — | |
| OpenVLA with TAPTBase Model=OpenVLA, Adaptation=TAPT, Deployment mode=VLA tool-family interface2026.05 | 82.4 | — | — | — | — | — | — | — | |
| AnchorRefine (GR-1)Size=0.22B2026.04 | 82.3 | — | — | — | — | — | — | — | |
| π0.5Base Model=π0.5, Adaptation=SFT, Deployment mode=Direct planner2026.05 | 80.4 | — | — | — | — | — | — | — | |
| OpenVLA-OFTBase Model=OpenVLA-OFT, Adaptation=SFT, Deployment mode=Direct planner2026.05 | 80.2 | — | — | — | — | — | — | — | |
| BlockVLAType=Block Diffusion, Training Steps=40k2026.05 | 78.6 | — | — | — | — | — | — | — | |
| OpenVLABase Model=OpenVLA, Adaptation=SFT, Deployment mode=Standalone2026.05 | 77.2 | — | — | — | — | — | — | — | |
| OpenVLABase Model=OpenVLA, Adaptation=SFT, Deployment mode=Direct planner2026.05 | 77 | — | — | — | — | — | — | — | |
| GR-1*Size=0.2B2026.04 | 74.5 | — | — | — | — | — | — | — | |
| GPT-5.4Paradigm=Agentic, Zero-shot=true, Shared System-1 executor=π0.52026.07 | 72 | — | — | — | — | — | — | — | |
| BlockVLAType=Block Diffusion, Training Steps=20k2026.05 | 70.2 | — | — | — | — | — | — | — | |
| BlockVLAType=Block Diffusion, Training Steps=30k2026.05 | 69.6 | — | — | — | — | — | — | — | |
| Qwen3-VL-8BParadigm=Agentic, Zero-shot=true, Shared System-1 executor=π0.52026.07 | 68 | — | — | — | — | — | — | — | |
| DitaSize=–2026.04 | 63.8 | — | — | — | — | — | — | — | |
| BlockVLAType=Block Diffusion, Training Steps=15k2026.05 | 62.2 | — | — | — | — | — | — | — | |
| RoboBrainParadigm=Agentic, Zero-shot=true, Shared System-1 executor=π0.52026.07 | 57 | — | — | — | — | — | — | — | |
| OpenVLASize=7B2026.04 | 53.7 | — | — | — | — | — | — | — | |
| OpenVLAType=Autoregressive2026.05 | 53.7 | — | — | — | — | — | — | — | |
| DDVLAType=Discrete Diffusion, Training Steps=50k2026.05 | 53.6 | — | — | — | — | — | — | — | |
| Diffusion PolicySize=–2026.04 | 50.5 | — | — | — | — | — | — | — | |
| BlockVLAType=Block Diffusion, Training Steps=10k2026.05 | 44.2 | — | — | — | — | — | — | — | |
| DDVLAType=Discrete Diffusion, Training Steps=40k2026.05 | 37.2 | — | — | — | — | — | — | — | |
| BlockVLAType=Block Diffusion, Training Steps=5k2026.05 | 35.8 | — | — | — | — | — | — | — | |
| DDVLAType=Discrete Diffusion, Training Steps=30k2026.05 | 23.2 | — | — | — | — | — | — | — | |
| DDVLAType=Discrete Diffusion, Training Steps=20k2026.05 | 18.6 | — | — | — | — | — | — | — | |
| DDVLAType=Discrete Diffusion, Training Steps=15k2026.05 | 3.8 | — | — | — | — | — | — | — | |
| DDVLAType=Discrete Diffusion, Training Steps=10k2026.05 | 3.6 | — | — | — | — | — | — | — | |
| DDVLAType=Discrete Diffusion, Training Steps=5k2026.05 | 0.4 | — | — | — | — | — | — | — | |
| CLaDInference time (Hz)=25, Memory (GB)=4, Parameters (B)=0.662026.03 | — | 25 | 4 | 0.66 | — | — | — | — | |
| OpenPI0.5Training Paradigm=OpenX + Web Co-train2026.05 | — | — | — | — | 92 | 76.1 | 65.6 | 47.7 | |
| OpenVLAInference time (Hz)=6, Memory (GB)=15, Parameters (B)=7.02026.03 | — | 6 | 15 | 7 | — | — | — | — | |
| OpenVLATraining Paradigm=OpenX Pretrain2026.05 | — | — | — | — | 55.5 | 20.5 | 12.4 | 7 | |
| OpenVLA-OFTTraining Paradigm=OpenX Pretrain2026.05 | — | — | — | — | 94.4 | 77.6 | 61.9 | 40.3 | |
| pi_0.5Inference time (Hz)=10, Memory (GB)=19, Parameters (B)=3.32026.03 | — | 10 | 19 | 3.3 | — | — | — | — | |
| StableVLATraining Paradigm=VLM Direct FT2026.05 | — | — | — | — | 93.6 | 76.3 | 62.4 | 45.3 | |
| VLA-Adapter-ProTraining Paradigm=VLM Direct FT2026.05 | — | — | — | — | 94.4 | 63.5 | 41 | 26.2 |