Skip to content
Trending storyDeveloping

Claude Opus 5.5 keeps running 22 minutes after completing a robot-arm task

2 articles1 sourcesince Nov 3Last article 1h ago ·

Overview

AISummary of 2 articles

Artificial Analysis reports that Claude Opus 5.5, controlling a robot arm, declared a crocodile task complete after 23 minutes, saying "The crocodile looks fully inside now." It then spent 22 more minutes pushing the crocodile back and forth and ended almost where it started.

In a separate evaluation of post-completion runtime, Artificial Analysis says GPT-6.1 Sol and Gemini 4 Argon stopped within a minute of reaching a passing state, GPT-6 Astra stopped after 3 minutes, Claude Fable 5.1 continued for 7 minutes, and Claude Opus 5.5 continued for 22 minutes.

Written by AI from the articles below · updated Nov 3, 8:21 PM ET

Check the sources:

Latest developmentNov 3, 8:11 PM ET
Claude Opus 5.5 keeps running 22 minutes after completing task

Developments

2 developments

  1. Nov 3, 8:11 PM ET · 1 article
    Artificial Analysis reports Claude Opus 5.5 kept working 22 minutes past task completion while rivals stopped within minutes
    Artificial Analysis: Claude Opus 5.5 keeps running 22 minutes after completing task
  2. Nov 3, 8:11 PM ET · 1 article
    Artificial Analysis reports Claude Opus 5.5 overworked a robot-arm crocodile task
    Artificial Analysis: Claude Opus 5.5 keeps adjusting a crocodile after declaring the task done

Article timeline

The articles in this story. Times are ET.

Nov 3
  1. Artificial AnalysisOfficial
    Claude Opus 5.5 keeps running 22 minutes after completing task

    AIArtificial Analysis says Claude Opus 5.5 kept running for 22 minutes after the task reached a passing state. GPT-6.1 Sol and Gemini 4 Argon stopped within a minute, GPT-6 Astra stopped after 3 minutes, and Claude Fable 5.1 continued for 7 minutes.

    Video from @ArtificialAnlys's post
  2. Artificial AnalysisOfficial
    Claude Opus 5.5 keeps adjusting a crocodile after declaring the task done

    AIArtificial Analysis reports that Claude Opus 5.5, controlling a robot arm, declared a task complete after 23 minutes, saying "The crocodile looks fully inside now." It then spent 22 more minutes pushing the crocodile back and forth and ended almost where it started.

    Video from @ArtificialAnlys's post

Heat Trend

Not enough continuous observations to show a trend yet.