artificial-intelligence

Google pitches Gemini 4 Argon for robots, but tests show limits

Google says its Gemini 4 Argon model can control robotic arms, scoring 87 percent on Google's test. An independent benchmark found such tasks take at least 30 minutes and cost ten times human labor.

By Standard Topic. AI-generated from the reporting listed under Coverage. Published . Updated .

This story is no longer being updated. It reflects reporting up to .

What happened

Google is promoting its Gemini 4 Argon model as a controller for robots, able to generate simulations from photos or sketches and evaluate training videos. Published example prompts show the demos still need substantial human guidance. The benchmarking firm Artificial Analysis then released AA-Robotics v0.1, the first evaluation of robot manipulation for general-purpose large language models.

Why it matters

The published example prompts show that robot control by a single AI model still depends heavily on human input. The independent benchmark finds such tasks are slow and expensive compared with human work.

Timeline

  1. Oct 15, 9:21 AM ET: Epoch AI scored Gemini 4 Argon at 165 on its Capabilities Index, tying Claude Fable 5.1 and trailing GPT-6 Astra and Claude Opus 5.5.
  2. Oct 15, 12:35 PM ET: Google said Gemini 4 Argon scores 87 percent on its own ERQA robotics benchmark, but its example prompts show the demos need a lot of human guidance.
  3. Oct 15, 7:55 PM ET: Artificial Analysis released AA-Robotics v0.1, which found robot arm tasks controlled by general-purpose LLMs take at least 30 minutes each and cost 10 times human labor.

Coverage

More on this story

More from artificial-intelligence

All artificial-intelligence news · artificial-intelligence story archive