Claude Opus 4.7 completes robotics setup tasks nearly 19 times faster than assisted humans
In Anthropic's controlled Project Fetch follow-up, Claude Opus 4.7 completed four robodog setup and programming tasks in about 9.6 minutes, nearly 19 times faster than the earlier Claude-assisted human team. The model still failed the final autonomous ball-retrieval task, and a researcher approved commands and task transitions.
0 comments · 0 votes
Sign in to join the discussion →
No comments yet. Start the discussion.
Why it moved the index
The physical evaluation shows a substantial practical gain in autonomous setup, sensing, and controller-programming tasks without human problem solving. The remaining command approvals, failed end task, and absence of low-level motor-policy learning materially limit the autonomy demonstrated.
Assessment history
-
R1
Toward 46 · confidence 86
Adds the missing physical robotics follow-up with its measured speedup, approval boundary, and failed final task.
15 Aug 2026
Share this page
-
DoomBench assesses “Claude Opus 4.7 completes robotics setup tasks nearly 19 times faster than assisted humans” as evidence moving toward doom, with magnitude 46 and confidence 86 out of 100 in the autonomy and agency category.
-
The DoomBench assessment of “Claude Opus 4.7 completes robotics setup tasks nearly 19 times faster than assisted humans” is based on reporting from Anthropic Research and records the editorial rationale, source quality, attribution, and...
-
DoomBench summarizes “Claude Opus 4.7 completes robotics setup tasks nearly 19 times faster than assisted humans” as follows: In Anthropic's controlled Project Fetch follow-up, Claude Opus 4.7 completed four robodog setup and...
https://www.doombench.com/news/claude-opus-4-7-completes-robotics-setup-tasks-nearly-19-times-faster-than-assisted-humans-2026-06-18