{
  "name": "AI Model Performance (policy scorecard)",
  "publisher": "Bits 2 Atoms",
  "url": "https://b2atoms.com/data/policy-scorecard/",
  "methodology": "https://b2atoms.com/data/methodology/",
  "checked": "23 Oct 2026",
  "grades": {
    "A": "Independent, with logs",
    "B": "Paper or report with method (rubric or error bars)",
    "C": "Claim; method unpublished"
  },
  "axes": [
    {
      "k": "succ",
      "name": "Success",
      "q": "Does it do the trained task?",
      "unit": "% of trials, trained setup"
    },
    {
      "k": "rob",
      "name": "Robustness",
      "q": "Does it survive the cell drifting?",
      "unit": "% under the perturbation set"
    },
    {
      "k": "lang",
      "name": "Instruction",
      "q": "Does rewording the instruction break it?",
      "unit": "% under paraphrase"
    },
    {
      "k": "rec",
      "name": "Recovery",
      "q": "Does it fix its own mistakes?",
      "unit": "% of injected failures recovered"
    },
    {
      "k": "speed",
      "name": "Speed",
      "q": "Is it as fast as a trained person?",
      "unit": "human cycle time ÷ robot cycle time"
    },
    {
      "k": "int",
      "name": "Interventions",
      "q": "How often does a person step in?",
      "unit": "per 1,000 cycles"
    }
  ],
  "rows": [
    {
      "model": "π0.7",
      "organisation": "Physical Intelligence",
      "weights": "closed",
      "axis": "Success",
      "status": "published",
      "value": 90,
      "value_display": "~90%",
      "approximate": true,
      "setting": "real robot",
      "grade": "B",
      "trials": null,
      "context": "Seen tasks",
      "note": "Paper: 'seen tasks often have success rates in excess of 90%'. Fig. 6: laundry ~95%, espresso ~95%, box building ~85%, hardest laundry item ~70%. Error bars and scoring rubrics published (App. G); trial counts not stated.",
      "source": "π0.7 paper (Physical Intelligence, PDF)",
      "source_url": "https://www.pi.website/download/pi07.pdf",
      "checked": "23 Oct 2026"
    },
    {
      "model": "π0.7",
      "organisation": "Physical Intelligence",
      "weights": "closed",
      "axis": "Robustness",
      "status": "published",
      "value": 70,
      "value_display": "~70%",
      "approximate": true,
      "setting": "real robot",
      "grade": "B",
      "trials": null,
      "context": "Unseen tasks / robots",
      "note": "'Unseen tasks or unseen task-robot combinations have success rates in the 60–80% range.' Closest available proxy; not a fixed perturbation set.",
      "source": "π0.7 paper (Physical Intelligence, PDF)",
      "source_url": "https://www.pi.website/download/pi07.pdf",
      "checked": "23 Oct 2026"
    },
    {
      "model": "π0.7",
      "organisation": "Physical Intelligence",
      "weights": "closed",
      "axis": "Instruction",
      "status": "published",
      "value": 87,
      "value_display": "~87%",
      "approximate": true,
      "setting": "real robot",
      "grade": "B",
      "trials": null,
      "context": "Unseen kitchens",
      "note": "Instruction-following success across 14 scenarios of 3–6 instructions: ~87% in 4 unseen kitchens, ~73% in 2 unseen bedrooms (read from chart). π0.5 scored ~28% on the same test.",
      "source": "π0.7 paper (Physical Intelligence, PDF)",
      "source_url": "https://www.pi.website/download/pi07.pdf",
      "checked": "23 Oct 2026"
    },
    {
      "model": "π0.7",
      "organisation": "Physical Intelligence",
      "weights": "closed",
      "axis": "Recovery",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": "Not reported as a rate. Trained on failure and correction data, but no recovery metric.",
      "source": "π0.7 paper (Physical Intelligence, PDF)",
      "source_url": "https://www.pi.website/download/pi07.pdf",
      "checked": "23 Oct 2026"
    },
    {
      "model": "π0.7",
      "organisation": "Physical Intelligence",
      "weights": "closed",
      "axis": "Speed",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": "No cycle time against a person. Throughput is normalised to RL specialist models (1.0–1.5×). The human comparison in App. F is on success: 80% vs expert teleoperators' 80.6% on zero-shot shirt folding.",
      "source": "π0.7 paper (Physical Intelligence, PDF)",
      "source_url": "https://www.pi.website/download/pi07.pdf",
      "checked": "23 Oct 2026"
    },
    {
      "model": "π0.7",
      "organisation": "Physical Intelligence",
      "weights": "closed",
      "axis": "Interventions",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": "Not reported.",
      "source": "π0.7 paper (Physical Intelligence, PDF)",
      "source_url": "https://www.pi.website/download/pi07.pdf",
      "checked": "23 Oct 2026"
    },
    {
      "model": "GEN-1",
      "organisation": "Generalist",
      "weights": "closed",
      "axis": "Success",
      "status": "published",
      "value": 99,
      "value_display": "~99%",
      "approximate": true,
      "setting": "real robot",
      "grade": "C",
      "trials": null,
      "context": "Selected tasks",
      "note": "'Average success rates to 99% on tasks where previous models achieved 64%.' Rubrics, trial counts and intervals not published.",
      "source": "GEN-1 launch coverage, The Robot Report",
      "source_url": "https://www.therobotreport.com/generalist-introduces-gen-1-general-purpose-model-for-physical-ai/#:~:text=99%25",
      "checked": "23 Oct 2026"
    },
    {
      "model": "GEN-1",
      "organisation": "Generalist",
      "weights": "closed",
      "axis": "Robustness",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": "Not reported.",
      "source": "GEN-1 launch coverage, The Robot Report",
      "source_url": "https://www.therobotreport.com/generalist-introduces-gen-1-general-purpose-model-for-physical-ai/",
      "checked": "23 Oct 2026"
    },
    {
      "model": "GEN-1",
      "organisation": "Generalist",
      "weights": "closed",
      "axis": "Instruction",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": "Not reported.",
      "source": "GEN-1 launch coverage, The Robot Report",
      "source_url": "https://www.therobotreport.com/generalist-introduces-gen-1-general-purpose-model-for-physical-ai/",
      "checked": "23 Oct 2026"
    },
    {
      "model": "GEN-1",
      "organisation": "Generalist",
      "weights": "closed",
      "axis": "Recovery",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": "Shown in demos (regrasping, switching hands), no rate published.",
      "source": "GEN-1 launch coverage, The Robot Report",
      "source_url": "https://www.therobotreport.com/generalist-introduces-gen-1-general-purpose-model-for-physical-ai/",
      "checked": "23 Oct 2026"
    },
    {
      "model": "GEN-1",
      "organisation": "Generalist",
      "weights": "closed",
      "axis": "Speed",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": "No human comparison. Box assembly in 12.1 s vs ~34 s for GEN-0 and π0 — against prior models, not a person.",
      "source": "GEN-1 launch coverage, The Robot Report",
      "source_url": "https://www.therobotreport.com/generalist-introduces-gen-1-general-purpose-model-for-physical-ai/#:~:text=12.1",
      "checked": "23 Oct 2026"
    },
    {
      "model": "GEN-1",
      "organisation": "Generalist",
      "weights": "closed",
      "axis": "Interventions",
      "status": "published",
      "value": "0 / 1,800+",
      "value_display": "0 / 1,800+",
      "approximate": false,
      "setting": "real robot",
      "grade": "C",
      "trials": null,
      "context": "Consecutive runs",
      "note": "Uninterrupted runs: 1,800+ block packings, 200+ box folds, 200+ vacuum servicings, 86 T-shirt folds (company claim).",
      "source": "GEN-1 launch coverage, The Robot Report",
      "source_url": "https://www.therobotreport.com/generalist-introduces-gen-1-general-purpose-model-for-physical-ai/#:~:text=1%2C800",
      "checked": "23 Oct 2026"
    },
    {
      "model": "Gemini Robotics 2",
      "organisation": "Google DeepMind",
      "weights": "closed",
      "axis": "Success",
      "status": "published",
      "value": 78.9,
      "value_display": "78.9%",
      "approximate": false,
      "setting": "real robot",
      "grade": "C",
      "trials": null,
      "context": "Tool kitting · Franka",
      "note": "Per-task figures, trial counts not published. Precise insertion 89.6%, general pick-and-place 74.2%; whole-body pick from floor 45.7%.",
      "source": "Gemini Robotics 2, DeepMind blog",
      "source_url": "https://deepmind.google/blog/gemini-robotics-2-brings-whole-body-intelligence-to-robots/",
      "checked": "23 Oct 2026"
    },
    {
      "model": "Gemini Robotics 2",
      "organisation": "Google DeepMind",
      "weights": "closed",
      "axis": "Robustness",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": "Not reported for Gemini Robotics 2.",
      "source": "Gemini Robotics 2, DeepMind blog",
      "source_url": "https://deepmind.google/blog/gemini-robotics-2-brings-whole-body-intelligence-to-robots/",
      "checked": "23 Oct 2026"
    },
    {
      "model": "Gemini Robotics 2",
      "organisation": "Google DeepMind",
      "weights": "closed",
      "axis": "Instruction",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": "Not reported for Gemini Robotics 2.",
      "source": "Gemini Robotics 2, DeepMind blog",
      "source_url": "https://deepmind.google/blog/gemini-robotics-2-brings-whole-body-intelligence-to-robots/",
      "checked": "23 Oct 2026"
    },
    {
      "model": "Gemini Robotics 2",
      "organisation": "Google DeepMind",
      "weights": "closed",
      "axis": "Recovery",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": "Not reported.",
      "source": "Gemini Robotics 2, DeepMind blog",
      "source_url": "https://deepmind.google/blog/gemini-robotics-2-brings-whole-body-intelligence-to-robots/",
      "checked": "23 Oct 2026"
    },
    {
      "model": "Gemini Robotics 2",
      "organisation": "Google DeepMind",
      "weights": "closed",
      "axis": "Speed",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": "Not quantified. DeepMind says robots 'have more to advance in movement speed'.",
      "source": "Gemini Robotics 2, DeepMind blog",
      "source_url": "https://deepmind.google/blog/gemini-robotics-2-brings-whole-body-intelligence-to-robots/#:~:text=movement%20speed",
      "checked": "23 Oct 2026"
    },
    {
      "model": "Gemini Robotics 2",
      "organisation": "Google DeepMind",
      "weights": "closed",
      "axis": "Interventions",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": "Not reported.",
      "source": "Gemini Robotics 2, DeepMind blog",
      "source_url": "https://deepmind.google/blog/gemini-robotics-2-brings-whole-body-intelligence-to-robots/",
      "checked": "23 Oct 2026"
    },
    {
      "model": "π0.5",
      "organisation": "Physical Intelligence",
      "weights": "open",
      "axis": "Success",
      "status": "published",
      "value": 56.25,
      "value_display": "56.3%",
      "approximate": false,
      "setting": "real robot",
      "grade": "B",
      "trials": 80,
      "context": "SO-101 · 4 tasks",
      "note": "Average of pen transfer 95%, colour sorting 10%, packing 55%, pen placement 65%; 100 demos per task. π0.5's own paper reports 60–80% task progress in unseen homes (Fig. 7b, 10 trials per task per home).",
      "source": "SO-101 VLA benchmark, arXiv 2606.08881",
      "source_url": "https://arxiv.org/html/2606.08881v2#:~:text=56.25",
      "checked": "23 Oct 2026"
    },
    {
      "model": "π0.5",
      "organisation": "Physical Intelligence",
      "weights": "open",
      "axis": "Robustness",
      "status": "published",
      "value": 85.31,
      "value_display": "85.3%",
      "approximate": false,
      "setting": "simulation",
      "grade": "B",
      "trials": null,
      "context": "LIBERO-Plus",
      "note": "Non-language perturbations (camera, robot state, lighting, layout), simulation.",
      "source": "IndustrialVLA-Bench, arXiv 2609.25562",
      "source_url": "https://arxiv.org/html/2609.25562v2#:~:text=85.31",
      "checked": "23 Oct 2026"
    },
    {
      "model": "π0.5",
      "organisation": "Physical Intelligence",
      "weights": "open",
      "axis": "Instruction",
      "status": "published",
      "value": 71.33,
      "value_display": "71.3%",
      "approximate": false,
      "setting": "simulation",
      "grade": "B",
      "trials": null,
      "context": "LIBERO-Para",
      "note": "Paraphrased instructions, 25.2 points below clean. Real-world: π0.5 paper Fig. 9 shows ~85% language following on familiar objects, ~70% on new ones (read from chart).",
      "source": "IndustrialVLA-Bench, arXiv 2609.25562",
      "source_url": "https://arxiv.org/html/2609.25562v2#:~:text=71.33",
      "checked": "23 Oct 2026"
    },
    {
      "model": "π0.5",
      "organisation": "Physical Intelligence",
      "weights": "open",
      "axis": "Recovery",
      "status": "published",
      "value": 30.77,
      "value_display": "30.8%",
      "approximate": false,
      "setting": "real robot",
      "grade": "B",
      "trials": null,
      "context": "SO-101",
      "note": "Share of failures the policy recovered from without help.",
      "source": "SO-101 VLA benchmark, arXiv 2606.08881",
      "source_url": "https://arxiv.org/html/2606.08881v2#:~:text=30.77",
      "checked": "23 Oct 2026"
    },
    {
      "model": "π0.5",
      "organisation": "Physical Intelligence",
      "weights": "open",
      "axis": "Speed",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": "No cycle time against a person. Related, not equivalent: 58.8 ms per policy call (Table 5). Home tasks take 2–5 minutes, no human comparison.",
      "source": "IndustrialVLA-Bench, arXiv 2609.25562",
      "source_url": "https://arxiv.org/html/2609.25562v2#:~:text=58.80",
      "checked": "23 Oct 2026"
    },
    {
      "model": "π0.5",
      "organisation": "Physical Intelligence",
      "weights": "open",
      "axis": "Interventions",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": "Not reported. The π0.5 paper scores task progress, not interventions.",
      "source": "π0.5 paper, arXiv 2504.16054",
      "source_url": "https://arxiv.org/html/2504.16054",
      "checked": "23 Oct 2026"
    },
    {
      "model": "π0",
      "organisation": "Physical Intelligence",
      "weights": "open",
      "axis": "Success",
      "status": "published",
      "value": 72.25,
      "value_display": "72.3%",
      "approximate": false,
      "setting": "real robot",
      "grade": "B",
      "trials": 200,
      "context": "ALOHA · 4 tasks",
      "note": "Clean dish 87%, sponge in pot 83%, unzip bag 67%, fold shorts 52% (200 demos); 50 trials per task.",
      "source": "Benchmarking VLAs on ALOHA, arXiv 2511.11298",
      "source_url": "https://arxiv.org/html/2511.11298v1",
      "checked": "23 Oct 2026"
    },
    {
      "model": "π0",
      "organisation": "Physical Intelligence",
      "weights": "open",
      "axis": "Robustness",
      "status": "published",
      "value": 58.5,
      "value_display": "58.5%",
      "approximate": false,
      "setting": "real robot",
      "grade": "B",
      "trials": 200,
      "context": "ALOHA · unseen objects",
      "note": "Unseen objects at random positions: 81%, 73%, 37%, 43% — 13.8 points below in-distribution. Highest robustness of any model with real trial counts.",
      "source": "Benchmarking VLAs on ALOHA, arXiv 2511.11298",
      "source_url": "https://arxiv.org/html/2511.11298v1",
      "checked": "23 Oct 2026"
    },
    {
      "model": "π0",
      "organisation": "Physical Intelligence",
      "weights": "open",
      "axis": "Instruction",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": null,
      "source": null,
      "source_url": null,
      "checked": "23 Oct 2026"
    },
    {
      "model": "π0",
      "organisation": "Physical Intelligence",
      "weights": "open",
      "axis": "Recovery",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": null,
      "source": null,
      "source_url": null,
      "checked": "23 Oct 2026"
    },
    {
      "model": "π0",
      "organisation": "Physical Intelligence",
      "weights": "open",
      "axis": "Speed",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": "No human comparison. Time to success 13–85 s across successful trials.",
      "source": "Benchmarking VLAs on ALOHA, arXiv 2511.11298",
      "source_url": "https://arxiv.org/html/2511.11298v1",
      "checked": "23 Oct 2026"
    },
    {
      "model": "π0",
      "organisation": "Physical Intelligence",
      "weights": "open",
      "axis": "Interventions",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": null,
      "source": null,
      "source_url": null,
      "checked": "23 Oct 2026"
    },
    {
      "model": "GR00T N1.7",
      "organisation": "NVIDIA",
      "weights": "open",
      "axis": "Success",
      "status": "published",
      "value": 97.88,
      "value_display": "97.9%",
      "approximate": false,
      "setting": "simulation",
      "grade": "B",
      "trials": null,
      "context": "LIBERO clean",
      "note": "Clean LIBERO suite, simulation. No published real-robot success rate for N1.7 itself.",
      "source": "IndustrialVLA-Bench, arXiv 2609.25562",
      "source_url": "https://arxiv.org/html/2609.25562v2#:~:text=97.88",
      "checked": "23 Oct 2026"
    },
    {
      "model": "GR00T N1.7",
      "organisation": "NVIDIA",
      "weights": "open",
      "axis": "Robustness",
      "status": "published",
      "value": 75.12,
      "value_display": "75.1%",
      "approximate": false,
      "setting": "simulation",
      "grade": "B",
      "trials": null,
      "context": "LIBERO-Plus",
      "note": "Non-language perturbations, simulation.",
      "source": "IndustrialVLA-Bench, arXiv 2609.25562",
      "source_url": "https://arxiv.org/html/2609.25562v2#:~:text=75.12",
      "checked": "23 Oct 2026"
    },
    {
      "model": "GR00T N1.7",
      "organisation": "NVIDIA",
      "weights": "open",
      "axis": "Instruction",
      "status": "published",
      "value": 74.26,
      "value_display": "74.3%",
      "approximate": false,
      "setting": "simulation",
      "grade": "B",
      "trials": null,
      "context": "LIBERO-Para",
      "note": "Paraphrased instructions, 23.6 points below clean.",
      "source": "IndustrialVLA-Bench, arXiv 2609.25562",
      "source_url": "https://arxiv.org/html/2609.25562v2#:~:text=74.26",
      "checked": "23 Oct 2026"
    },
    {
      "model": "GR00T N1.7",
      "organisation": "NVIDIA",
      "weights": "open",
      "axis": "Recovery",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": "Not reported.",
      "source": "Isaac GR00T repository (N1.7)",
      "source_url": "https://github.com/NVIDIA/Isaac-GR00T",
      "checked": "23 Oct 2026"
    },
    {
      "model": "GR00T N1.7",
      "organisation": "NVIDIA",
      "weights": "open",
      "axis": "Speed",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": "No cycle time against a person. Related, not equivalent: 19.1 ms per policy call (Table 5).",
      "source": "IndustrialVLA-Bench, arXiv 2609.25562",
      "source_url": "https://arxiv.org/html/2609.25562v2#:~:text=19.13",
      "checked": "23 Oct 2026"
    },
    {
      "model": "GR00T N1.7",
      "organisation": "NVIDIA",
      "weights": "open",
      "axis": "Interventions",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": "Not reported.",
      "source": "Isaac GR00T repository (N1.7)",
      "source_url": "https://github.com/NVIDIA/Isaac-GR00T",
      "checked": "23 Oct 2026"
    },
    {
      "model": "Cosmos Policy",
      "organisation": "NVIDIA",
      "weights": "open",
      "axis": "Success",
      "status": "published",
      "value": 97.63,
      "value_display": "97.6%",
      "approximate": false,
      "setting": "simulation",
      "grade": "B",
      "trials": null,
      "context": "LIBERO clean",
      "note": "Independent re-run, simulation. Its own paper reports 93.6% on real bimanual tasks and 98.5% on LIBERO.",
      "source": "IndustrialVLA-Bench, arXiv 2609.25562",
      "source_url": "https://arxiv.org/html/2609.25562v2#:~:text=97.63",
      "checked": "23 Oct 2026"
    },
    {
      "model": "Cosmos Policy",
      "organisation": "NVIDIA",
      "weights": "open",
      "axis": "Robustness",
      "status": "published",
      "value": 80.47,
      "value_display": "80.5%",
      "approximate": false,
      "setting": "simulation",
      "grade": "B",
      "trials": null,
      "context": "LIBERO-Plus",
      "note": "Non-language perturbations, simulation.",
      "source": "IndustrialVLA-Bench, arXiv 2609.25562",
      "source_url": "https://arxiv.org/html/2609.25562v2#:~:text=80.47",
      "checked": "23 Oct 2026"
    },
    {
      "model": "Cosmos Policy",
      "organisation": "NVIDIA",
      "weights": "open",
      "axis": "Instruction",
      "status": "published",
      "value": 70.5,
      "value_display": "70.5%",
      "approximate": false,
      "setting": "simulation",
      "grade": "B",
      "trials": null,
      "context": "LIBERO-Para",
      "note": "Paraphrased instructions, 27.1 points below clean.",
      "source": "IndustrialVLA-Bench, arXiv 2609.25562",
      "source_url": "https://arxiv.org/html/2609.25562v2#:~:text=70.50",
      "checked": "23 Oct 2026"
    },
    {
      "model": "Cosmos Policy",
      "organisation": "NVIDIA",
      "weights": "open",
      "axis": "Recovery",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": null,
      "source": null,
      "source_url": null,
      "checked": "23 Oct 2026"
    },
    {
      "model": "Cosmos Policy",
      "organisation": "NVIDIA",
      "weights": "open",
      "axis": "Speed",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": "No human comparison. 21.3 ms per policy call (Table 5).",
      "source": "IndustrialVLA-Bench, arXiv 2609.25562",
      "source_url": "https://arxiv.org/html/2609.25562v2#:~:text=21.26",
      "checked": "23 Oct 2026"
    },
    {
      "model": "Cosmos Policy",
      "organisation": "NVIDIA",
      "weights": "open",
      "axis": "Interventions",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": null,
      "source": null,
      "source_url": null,
      "checked": "23 Oct 2026"
    },
    {
      "model": "UnifoLM-VLA-0",
      "organisation": "Unitree",
      "weights": "open",
      "axis": "Success",
      "status": "published",
      "value": 97.92,
      "value_display": "97.9%",
      "approximate": false,
      "setting": "simulation",
      "grade": "B",
      "trials": null,
      "context": "LIBERO clean",
      "note": "Clean LIBERO, simulation.",
      "source": "IndustrialVLA-Bench, arXiv 2609.25562",
      "source_url": "https://arxiv.org/html/2609.25562v2#:~:text=97.92",
      "checked": "23 Oct 2026"
    },
    {
      "model": "UnifoLM-VLA-0",
      "organisation": "Unitree",
      "weights": "open",
      "axis": "Robustness",
      "status": "published",
      "value": 78.78,
      "value_display": "78.8%",
      "approximate": false,
      "setting": "simulation",
      "grade": "B",
      "trials": null,
      "context": "LIBERO-Plus",
      "note": "Non-language perturbations, simulation.",
      "source": "IndustrialVLA-Bench, arXiv 2609.25562",
      "source_url": "https://arxiv.org/html/2609.25562v2#:~:text=78.78",
      "checked": "23 Oct 2026"
    },
    {
      "model": "UnifoLM-VLA-0",
      "organisation": "Unitree",
      "weights": "open",
      "axis": "Instruction",
      "status": "published",
      "value": 82.24,
      "value_display": "82.2%",
      "approximate": false,
      "setting": "simulation",
      "grade": "B",
      "trials": null,
      "context": "LIBERO-Para",
      "note": "Most robust to rephrasing of the six models tested: 15.7 points below clean.",
      "source": "IndustrialVLA-Bench, arXiv 2609.25562",
      "source_url": "https://arxiv.org/html/2609.25562v2#:~:text=82.24",
      "checked": "23 Oct 2026"
    },
    {
      "model": "UnifoLM-VLA-0",
      "organisation": "Unitree",
      "weights": "open",
      "axis": "Recovery",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": null,
      "source": null,
      "source_url": null,
      "checked": "23 Oct 2026"
    },
    {
      "model": "UnifoLM-VLA-0",
      "organisation": "Unitree",
      "weights": "open",
      "axis": "Speed",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": "No human comparison. 19.8 ms per policy call (Table 5).",
      "source": "IndustrialVLA-Bench, arXiv 2609.25562",
      "source_url": "https://arxiv.org/html/2609.25562v2#:~:text=19.82",
      "checked": "23 Oct 2026"
    },
    {
      "model": "UnifoLM-VLA-0",
      "organisation": "Unitree",
      "weights": "open",
      "axis": "Interventions",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": null,
      "source": null,
      "source_url": null,
      "checked": "23 Oct 2026"
    },
    {
      "model": "Xiaomi-Robotics-0",
      "organisation": "Xiaomi",
      "weights": "open",
      "axis": "Success",
      "status": "published",
      "value": 98.1,
      "value_display": "98.1%",
      "approximate": false,
      "setting": "simulation",
      "grade": "B",
      "trials": null,
      "context": "LIBERO clean",
      "note": "Highest clean LIBERO score of the six models tested, simulation.",
      "source": "IndustrialVLA-Bench, arXiv 2609.25562",
      "source_url": "https://arxiv.org/html/2609.25562v2#:~:text=98.10",
      "checked": "23 Oct 2026"
    },
    {
      "model": "Xiaomi-Robotics-0",
      "organisation": "Xiaomi",
      "weights": "open",
      "axis": "Robustness",
      "status": "published",
      "value": 73.91,
      "value_display": "73.9%",
      "approximate": false,
      "setting": "simulation",
      "grade": "B",
      "trials": null,
      "context": "LIBERO-Plus",
      "note": "Non-language perturbations, simulation.",
      "source": "IndustrialVLA-Bench, arXiv 2609.25562",
      "source_url": "https://arxiv.org/html/2609.25562v2#:~:text=73.91",
      "checked": "23 Oct 2026"
    },
    {
      "model": "Xiaomi-Robotics-0",
      "organisation": "Xiaomi",
      "weights": "open",
      "axis": "Instruction",
      "status": "published",
      "value": 75.73,
      "value_display": "75.7%",
      "approximate": false,
      "setting": "simulation",
      "grade": "B",
      "trials": null,
      "context": "LIBERO-Para",
      "note": "Paraphrased instructions, 22.4 points below clean.",
      "source": "IndustrialVLA-Bench, arXiv 2609.25562",
      "source_url": "https://arxiv.org/html/2609.25562v2#:~:text=75.73",
      "checked": "23 Oct 2026"
    },
    {
      "model": "Xiaomi-Robotics-0",
      "organisation": "Xiaomi",
      "weights": "open",
      "axis": "Recovery",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": null,
      "source": null,
      "source_url": null,
      "checked": "23 Oct 2026"
    },
    {
      "model": "Xiaomi-Robotics-0",
      "organisation": "Xiaomi",
      "weights": "open",
      "axis": "Speed",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": "No human comparison. 25.4 ms per policy call (Table 5).",
      "source": "IndustrialVLA-Bench, arXiv 2609.25562",
      "source_url": "https://arxiv.org/html/2609.25562v2#:~:text=25.44",
      "checked": "23 Oct 2026"
    },
    {
      "model": "Xiaomi-Robotics-0",
      "organisation": "Xiaomi",
      "weights": "open",
      "axis": "Interventions",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": null,
      "source": null,
      "source_url": null,
      "checked": "23 Oct 2026"
    },
    {
      "model": "Fast-WAM",
      "organisation": "Academic (Yuan, Dong, Liu, Zhao)",
      "weights": "open",
      "axis": "Success",
      "status": "published",
      "value": 96.55,
      "value_display": "96.5%",
      "approximate": false,
      "setting": "simulation",
      "grade": "B",
      "trials": null,
      "context": "LIBERO clean",
      "note": "World action model, simulation.",
      "source": "IndustrialVLA-Bench, arXiv 2609.25562",
      "source_url": "https://arxiv.org/html/2609.25562v2#:~:text=96.55",
      "checked": "23 Oct 2026"
    },
    {
      "model": "Fast-WAM",
      "organisation": "Academic (Yuan, Dong, Liu, Zhao)",
      "weights": "open",
      "axis": "Robustness",
      "status": "published",
      "value": 70.69,
      "value_display": "70.7%",
      "approximate": false,
      "setting": "simulation",
      "grade": "B",
      "trials": null,
      "context": "LIBERO-Plus",
      "note": "Non-language perturbations, simulation.",
      "source": "IndustrialVLA-Bench, arXiv 2609.25562",
      "source_url": "https://arxiv.org/html/2609.25562v2#:~:text=70.69",
      "checked": "23 Oct 2026"
    },
    {
      "model": "Fast-WAM",
      "organisation": "Academic (Yuan, Dong, Liu, Zhao)",
      "weights": "open",
      "axis": "Instruction",
      "status": "published",
      "value": 51.16,
      "value_display": "51.2%",
      "approximate": false,
      "setting": "simulation",
      "grade": "B",
      "trials": null,
      "context": "LIBERO-Para",
      "note": "Weakest on rephrased instructions: 45.4 points below clean.",
      "source": "IndustrialVLA-Bench, arXiv 2609.25562",
      "source_url": "https://arxiv.org/html/2609.25562v2#:~:text=51.16",
      "checked": "23 Oct 2026"
    },
    {
      "model": "Fast-WAM",
      "organisation": "Academic (Yuan, Dong, Liu, Zhao)",
      "weights": "open",
      "axis": "Recovery",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": null,
      "source": null,
      "source_url": null,
      "checked": "23 Oct 2026"
    },
    {
      "model": "Fast-WAM",
      "organisation": "Academic (Yuan, Dong, Liu, Zhao)",
      "weights": "open",
      "axis": "Speed",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": null,
      "source": null,
      "source_url": null,
      "checked": "23 Oct 2026"
    },
    {
      "model": "Fast-WAM",
      "organisation": "Academic (Yuan, Dong, Liu, Zhao)",
      "weights": "open",
      "axis": "Interventions",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": null,
      "source": null,
      "source_url": null,
      "checked": "23 Oct 2026"
    },
    {
      "model": "Wall-X",
      "organisation": "X Square Robot",
      "weights": "open",
      "axis": "Success",
      "status": "published",
      "value": 51.25,
      "value_display": "51.3%",
      "approximate": false,
      "setting": "real robot",
      "grade": "B",
      "trials": 80,
      "context": "SO-101 · 4 tasks",
      "note": "Pen transfer 95%, colour sorting 0%, packing 30%, pen placement 80%; 20 trials per task.",
      "source": "SO-101 VLA benchmark, arXiv 2606.08881",
      "source_url": "https://arxiv.org/html/2606.08881v2#:~:text=51.25",
      "checked": "23 Oct 2026"
    },
    {
      "model": "Wall-X",
      "organisation": "X Square Robot",
      "weights": "open",
      "axis": "Robustness",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": null,
      "source": null,
      "source_url": null,
      "checked": "23 Oct 2026"
    },
    {
      "model": "Wall-X",
      "organisation": "X Square Robot",
      "weights": "open",
      "axis": "Instruction",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": null,
      "source": null,
      "source_url": null,
      "checked": "23 Oct 2026"
    },
    {
      "model": "Wall-X",
      "organisation": "X Square Robot",
      "weights": "open",
      "axis": "Recovery",
      "status": "published",
      "value": 20.51,
      "value_display": "20.5%",
      "approximate": false,
      "setting": "real robot",
      "grade": "B",
      "trials": null,
      "context": "SO-101",
      "note": "Share of failures recovered without help.",
      "source": "SO-101 VLA benchmark, arXiv 2606.08881",
      "source_url": "https://arxiv.org/html/2606.08881v2#:~:text=20.51",
      "checked": "23 Oct 2026"
    },
    {
      "model": "Wall-X",
      "organisation": "X Square Robot",
      "weights": "open",
      "axis": "Speed",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": null,
      "source": null,
      "source_url": null,
      "checked": "23 Oct 2026"
    },
    {
      "model": "Wall-X",
      "organisation": "X Square Robot",
      "weights": "open",
      "axis": "Interventions",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": null,
      "source": null,
      "source_url": null,
      "checked": "23 Oct 2026"
    },
    {
      "model": "SmolVLA",
      "organisation": "Hugging Face",
      "weights": "open",
      "axis": "Success",
      "status": "published",
      "value": 32.5,
      "value_display": "32.5%",
      "approximate": false,
      "setting": "real robot",
      "grade": "B",
      "trials": 80,
      "context": "SO-101 · 4 tasks",
      "note": "Independent benchmark. SmolVLA's own paper reports 78.3% on its own SO-100 tasks (vs 61.7% for π0).",
      "source": "SO-101 VLA benchmark, arXiv 2606.08881",
      "source_url": "https://arxiv.org/html/2606.08881v2#:~:text=32.5",
      "checked": "23 Oct 2026"
    },
    {
      "model": "SmolVLA",
      "organisation": "Hugging Face",
      "weights": "open",
      "axis": "Robustness",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": "Not reported. SmolVLA is not in LIBERO-Plus or IndustrialVLA-Bench.",
      "source": "IndustrialVLA-Bench, arXiv 2609.25562",
      "source_url": "https://arxiv.org/html/2609.25562v2",
      "checked": "23 Oct 2026"
    },
    {
      "model": "SmolVLA",
      "organisation": "Hugging Face",
      "weights": "open",
      "axis": "Instruction",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": "Not reported.",
      "source": "SmolVLA paper, arXiv 2506.01844",
      "source_url": "https://arxiv.org/html/2506.01844",
      "checked": "23 Oct 2026"
    },
    {
      "model": "SmolVLA",
      "organisation": "Hugging Face",
      "weights": "open",
      "axis": "Recovery",
      "status": "published",
      "value": 3.23,
      "value_display": "3.2%",
      "approximate": false,
      "setting": "real robot",
      "grade": "B",
      "trials": null,
      "context": "SO-101",
      "note": "Share of failures recovered without help.",
      "source": "SO-101 VLA benchmark, arXiv 2606.08881",
      "source_url": "https://arxiv.org/html/2606.08881v2#:~:text=3.23",
      "checked": "23 Oct 2026"
    },
    {
      "model": "SmolVLA",
      "organisation": "Hugging Face",
      "weights": "open",
      "axis": "Speed",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": "No human comparison. Its own paper: asynchronous inference finishes tasks 30% faster than synchronous (9.7 s vs 13.75 s) — against itself, not a person.",
      "source": "SmolVLA paper, arXiv 2506.01844",
      "source_url": "https://arxiv.org/html/2506.01844#:~:text=13.75",
      "checked": "23 Oct 2026"
    },
    {
      "model": "SmolVLA",
      "organisation": "Hugging Face",
      "weights": "open",
      "axis": "Interventions",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": "Not reported.",
      "source": "SmolVLA paper, arXiv 2506.01844",
      "source_url": "https://arxiv.org/html/2506.01844",
      "checked": "23 Oct 2026"
    },
    {
      "model": "OpenVLA-OFT",
      "organisation": "Stanford",
      "weights": "open",
      "axis": "Success",
      "status": "published",
      "value": 13,
      "value_display": "13%",
      "approximate": false,
      "setting": "real robot",
      "grade": "B",
      "trials": 200,
      "context": "ALOHA · 4 tasks",
      "note": "Independent test: clean dish 29%, sponge 23%, unzip 0%, fold 0%. Its own paper reports 97.1% on LIBERO (simulation).",
      "source": "Benchmarking VLAs on ALOHA, arXiv 2511.11298",
      "source_url": "https://arxiv.org/html/2511.11298v1",
      "checked": "23 Oct 2026"
    },
    {
      "model": "OpenVLA-OFT",
      "organisation": "Stanford",
      "weights": "open",
      "axis": "Robustness",
      "status": "published",
      "value": 6,
      "value_display": "6%",
      "approximate": false,
      "setting": "real robot",
      "grade": "B",
      "trials": 200,
      "context": "ALOHA · unseen objects",
      "note": "Unseen objects at random positions: 13%, 11%, 0%, 0%.",
      "source": "Benchmarking VLAs on ALOHA, arXiv 2511.11298",
      "source_url": "https://arxiv.org/html/2511.11298v1",
      "checked": "23 Oct 2026"
    },
    {
      "model": "OpenVLA-OFT",
      "organisation": "Stanford",
      "weights": "open",
      "axis": "Instruction",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": null,
      "source": null,
      "source_url": null,
      "checked": "23 Oct 2026"
    },
    {
      "model": "OpenVLA-OFT",
      "organisation": "Stanford",
      "weights": "open",
      "axis": "Recovery",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": null,
      "source": null,
      "source_url": null,
      "checked": "23 Oct 2026"
    },
    {
      "model": "OpenVLA-OFT",
      "organisation": "Stanford",
      "weights": "open",
      "axis": "Speed",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": null,
      "source": null,
      "source_url": null,
      "checked": "23 Oct 2026"
    },
    {
      "model": "OpenVLA-OFT",
      "organisation": "Stanford",
      "weights": "open",
      "axis": "Interventions",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": null,
      "source": null,
      "source_url": null,
      "checked": "23 Oct 2026"
    },
    {
      "model": "RDT-1B",
      "organisation": "Tsinghua University",
      "weights": "open",
      "axis": "Success",
      "status": "published",
      "value": 13.5,
      "value_display": "13.5%",
      "approximate": false,
      "setting": "real robot",
      "grade": "B",
      "trials": 200,
      "context": "ALOHA · 4 tasks",
      "note": "Clean dish 27%, sponge 24%, unzip 3%, fold 0%; 50 trials per task.",
      "source": "Benchmarking VLAs on ALOHA, arXiv 2511.11298",
      "source_url": "https://arxiv.org/html/2511.11298v1",
      "checked": "23 Oct 2026"
    },
    {
      "model": "RDT-1B",
      "organisation": "Tsinghua University",
      "weights": "open",
      "axis": "Robustness",
      "status": "published",
      "value": 6.25,
      "value_display": "6.3%",
      "approximate": false,
      "setting": "real robot",
      "grade": "B",
      "trials": 200,
      "context": "ALOHA · unseen objects",
      "note": "Unseen objects at random positions: 12%, 12%, 1%, 0%.",
      "source": "Benchmarking VLAs on ALOHA, arXiv 2511.11298",
      "source_url": "https://arxiv.org/html/2511.11298v1",
      "checked": "23 Oct 2026"
    },
    {
      "model": "RDT-1B",
      "organisation": "Tsinghua University",
      "weights": "open",
      "axis": "Instruction",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": null,
      "source": null,
      "source_url": null,
      "checked": "23 Oct 2026"
    },
    {
      "model": "RDT-1B",
      "organisation": "Tsinghua University",
      "weights": "open",
      "axis": "Recovery",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": null,
      "source": null,
      "source_url": null,
      "checked": "23 Oct 2026"
    },
    {
      "model": "RDT-1B",
      "organisation": "Tsinghua University",
      "weights": "open",
      "axis": "Speed",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": null,
      "source": null,
      "source_url": null,
      "checked": "23 Oct 2026"
    },
    {
      "model": "RDT-1B",
      "organisation": "Tsinghua University",
      "weights": "open",
      "axis": "Interventions",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": null,
      "source": null,
      "source_url": null,
      "checked": "23 Oct 2026"
    },
    {
      "model": "ACT",
      "organisation": "Stanford (ALOHA) · baseline, not a foundation model",
      "weights": "open",
      "axis": "Success",
      "status": "published",
      "value": 48.25,
      "value_display": "48.3%",
      "approximate": false,
      "setting": "real robot",
      "grade": "B",
      "trials": 200,
      "context": "ALOHA · 4 tasks",
      "note": "Clean dish 86%, sponge 85%, unzip 22%, fold 0%. On the SO-101 benchmark it averaged 33.75%.",
      "source": "Benchmarking VLAs on ALOHA, arXiv 2511.11298",
      "source_url": "https://arxiv.org/html/2511.11298v1",
      "checked": "23 Oct 2026"
    },
    {
      "model": "ACT",
      "organisation": "Stanford (ALOHA) · baseline, not a foundation model",
      "weights": "open",
      "axis": "Robustness",
      "status": "published",
      "value": 5.5,
      "value_display": "5.5%",
      "approximate": false,
      "setting": "real robot",
      "grade": "B",
      "trials": 200,
      "context": "ALOHA · unseen objects",
      "note": "Collapses off its training setup: 42.8 points below in-distribution.",
      "source": "Benchmarking VLAs on ALOHA, arXiv 2511.11298",
      "source_url": "https://arxiv.org/html/2511.11298v1",
      "checked": "23 Oct 2026"
    },
    {
      "model": "ACT",
      "organisation": "Stanford (ALOHA) · baseline, not a foundation model",
      "weights": "open",
      "axis": "Instruction",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": null,
      "source": null,
      "source_url": null,
      "checked": "23 Oct 2026"
    },
    {
      "model": "ACT",
      "organisation": "Stanford (ALOHA) · baseline, not a foundation model",
      "weights": "open",
      "axis": "Recovery",
      "status": "published",
      "value": 6.45,
      "value_display": "6.5%",
      "approximate": false,
      "setting": "real robot",
      "grade": "B",
      "trials": null,
      "context": "SO-101",
      "note": "Share of failures recovered without help.",
      "source": "SO-101 VLA benchmark, arXiv 2606.08881",
      "source_url": "https://arxiv.org/html/2606.08881v2#:~:text=6.45",
      "checked": "23 Oct 2026"
    },
    {
      "model": "ACT",
      "organisation": "Stanford (ALOHA) · baseline, not a foundation model",
      "weights": "open",
      "axis": "Speed",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": null,
      "source": null,
      "source_url": null,
      "checked": "23 Oct 2026"
    },
    {
      "model": "ACT",
      "organisation": "Stanford (ALOHA) · baseline, not a foundation model",
      "weights": "open",
      "axis": "Interventions",
      "status": "owed",
      "value": null,
      "value_display": null,
      "approximate": null,
      "setting": null,
      "grade": null,
      "trials": null,
      "context": null,
      "note": null,
      "source": null,
      "source_url": null,
      "checked": "23 Oct 2026"
    }
  ]
}