Anthropic tested Claude Opus 4.7 programming a robodog, achieving 20x speed improvement over human team with Opus 4.1.

Archived signal · Frontier Models

Anthropic@AnthropicAIx

Anthropic tested Claude Opus 4.7 programming a robodog, achieving 20x speed improvement over human team with Opus 4.1.

Claude Opus 4.7 autonomously programmed a robodog ~20x faster than the best human team aided by Opus 4.1.
The robodog still failed to fetch a beach ball, indicating limitations in physical task completion.

Where this lives now

This signal is part of the DigestOps evidence archive. The living answers built on evidence like this are at Frontier Models; agents can query it as JSON at /api/v1/cards/2 or over MCP.