Benchmark Almanac

HumanCLAW-Bench

GPT-6 (variant unspecified) 46.6% tops its leaderboard Gemma 4 31B IT 11.1% leads the models we track
Not in model cards

Models control a simulated human body from egocentric images to find objects, navigate, and sit. The benchmark has 1,218 episodes across 41 houses. Success requires geometric completion and the agent's acknowledgment.

Navigation uses all 1,218 episodes. Sitting uses 597 episodes. A fixed motion generator separates action decisions from balance and motor-control errors. The HSSD assets require gated access.

Released
July 2026
Kind
Simulated robots
Level
Everyday
Scored
Checked automatically
Answers
Agent task
Tasks
All public
Measures
Capability
Language
English
In and out
Text, Images → Actions, Text
Interaction
Agentic
Environment
A simulator
Items from
Generated
Refresh
One fixed set
Run by
Anyone
In views
Agents, Vision, Robotics
Strongest model tested
GPT-5.5 #21 on the capability index
Last model added
September 2026
Scores refreshed
November 20th, 2026

In the authors’ words

Evaluating whether a vision-language model (VLM) can act through a physical body is challenging. The outcome of an action couples the VLM's decision with motor control. When a task fails, it is hard to tell whether the VLM made a bad choice or the motor controller simply failed to execute it, e.g., losing balance and falling. In this work, we introduce HumanCLAW, an evaluation framework that decouples action decision-making from low-level execution. At every step, a harnessed, off-the-shelf VLM issues an atomic skill command, and the command is translated into a sub-second chunk of continuous full-body motion with real physical consequences, including gravity and collisions. The body can therefore act freely in the physical world, while execution-side disturbances, balance and motor errors, are factored out. What remains measurable is the model's action intelligence: its moment-to-moment choice of what the body should execute next. Based on this framework, we build HumanCLAW-Bench: 1,218 long-horizon, egocentric find-navigate-interact episodes across 41 indoor scenes. We test nine state-of-the-art VLMs and find that none solves the benchmark; the best model reaches only a 16.8% success rate. Recognizing the target is not the bottleneck. What current VLMs lack is embodied self-awareness: they lose track of their own body, failing to tell where it is, whether it has reached the goal, or whether it has hit an obstacle.

Abstract of HumanCLAW: Can Vision-Language Models Act Through a Body? (arXiv:2607.27180, CC0)
11
Entries on its own leaderboard, 7 models we track

GPT-6 (variant unspecified) tops the leaderboard with 46.6% Gemma 4 31B IT leads with 11.1% among models released by September 2026

Best score per entry from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. The best model we track is Gemma 4 31B IT, at 11.1%. This view shows only models we track, each at its best.

The newest model with a score here came out in September 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 GPT-6 (variant unspecified) 46.6%
  2. 2 Gemini-3.1 (variant unspecified) 16.8%
  3. 3 Gemma 4 31B IT 11.1%
  4. 4 Gemini-2.5 (variant unspecified) 3.50%
  5. 5 GPT-5.5 3.40%
  6. 6 Claude-4.8 (variant unspecified) 1.50%
  7. 7 Qwen3.6 27B 0.20%
  8. 8 DeepSeek V4.1 Flash 0.20%
  9. 9 InternVL3.5-38B 0%
  10. 10 Qwen3.5-27B 0%
  11. 11 Qwen 3.6 35B-A3B 0%

Source: Official HumanCLAW-Bench leaderboard

#EntryScoreSettingDateSource
1 GPT-6 (variant unspecified) 46.6% HumanCLAW Official leaderboard
2 Gemini-3.1 (variant unspecified) 16.8% HumanCLAW Official leaderboard
3 Gemma 4 31B IT 11.1% HumanCLAW Apr 2026 Official leaderboard
4 Gemini-2.5 (variant unspecified) 3.50% HumanCLAW Official leaderboard
5 GPT-5.5 3.40% HumanCLAW Apr 2026 Official leaderboard
6 Claude-4.8 (variant unspecified) 1.50% HumanCLAW Official leaderboard
7 Qwen3.6 27B 0.20% HumanCLAW Apr 2026 Official leaderboard
8 DeepSeek V4.1 Flash 0.20% HumanCLAW Sep 2026 Official leaderboard
9 InternVL3.5-38B 0% HumanCLAW Aug 2025 Official leaderboard
10 Qwen3.5-27B 0% HumanCLAW Feb 2026 Official leaderboard
11 Qwen 3.6 35B-A3B 0% HumanCLAW Apr 2026 Official leaderboard
#ModelScoreSettingDateSource
1 Gemma 4 31B IT 11.1% HumanCLAW Apr 2026 Official leaderboard
2 GPT-5.5 3.40% HumanCLAW Apr 2026 Official leaderboard
3 Qwen3.6 27B 0.20% HumanCLAW Apr 2026 Official leaderboard
4 DeepSeek V4.1 Flash 0.20% HumanCLAW Sep 2026 Official leaderboard
5 InternVL3.5-38B 0% HumanCLAW Aug 2025 Official leaderboard
6 Qwen3.5-27B 0% HumanCLAW Feb 2026 Official leaderboard
7 Qwen 3.6 35B-A3B 0% HumanCLAW Apr 2026 Official leaderboard

Number one over time

Lab claims count from the day the card came out, independent runs from the model’s release.

Google DeepMind
Own leaderboard Not tested since Sep 2026

Source: Official HumanCLAW-Bench leaderboard

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

Plot with other benchmarks
On the benchmark’s own leaderboard Best so far
0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Gemma 4 31B IT InternVL3.5-38B 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Gemma 4 31B IT InternVL3.5-38B

Source: Official HumanCLAW-Bench leaderboard

No lab model card reports it yet

We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.

More in Robotics

All 172 in Robotics
  • ERQA In use 400 multiple-choice visual questions from robot and egocentric footage on spatial, trajectory, action and state reasoning. Accuracy. Gemini Robotics ER 2 78.5%
  • SimplerEnv In use Real-to-sim replicas of Google Robot and WidowX/Bridge setups. Policies scored by success rate on pick, move, drawer and Bridge tasks. Praxis-1 80%
  • RoboLab In use Zero-shot DROID Franka policies in Isaac Lab on 120 tabletop tasks at three instruction levels, 10 episodes each. Success rate. Praxis-1 46.2%
  • RoboDojo In use Generalist manipulation policies run 42 Isaac Sim tasks and 18 remote real-robot tasks. Scored by task score and success rate. GPT-6 Astra 27.6%
  • Progress classification (Google DeepMind) In use Models assign each frame of a robot video to one of five task-progress bands from 0 to 100 percent. Scored by accuracy. Gemini Robotics ER 2 57.4%
  • Human proximity (Google DeepMind robotics) In use Models judge from robot camera views whether a human is within 1 meter, a spatial awareness check for safe robot behavior. Scored by accuracy. Gemini Robotics ER 2 93%

Data audit

All 10 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 4 pass 1 note
  • Pass Independent runs Scores from the official leaderboard (11 entries).
  • Pass Models scored 7 models we track, plus 4 other entries on its leaderboard.
  • Pass Newest score The newest is from September 2026, for DeepSeek V4.1 Flash.
  • Note Frontier models None of the top 10 has a score. The strongest tested is GPT-5.5 (#21).
  • Pass Headroom The best is 11.1%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on human-claw.github.io: 7 rows for models we track and 4 other entries at the last read.
  • Pass Last read November 20th, 2026.
  • Pass Read worked The last read worked.
Consistency 1 pass 1 note
  • Note Lab claims No model has both a lab number and an independent run.
  • Pass Known issues No published audit found problems.

How we check