Skip to content
NewA calmer, faster ZeroHour

Gemini 4 Carbon delays code edits after finding the bug, blog says

Original title: Why is Gemini 4 Carbon taking so long?

Hacker News · AIby deenadz ()1 min read2
AI summary · x-ai/grok-4.7Summaries by Vani AI

A check-in every five turns cut Carbon's wait after naming the cause from about 25 minutes to 10.

A kernel.dev blog compared GPT-6 Astra at max effort with Gemini 4 Carbon at high on a sample debugging trial. Both models named the cause early, but Carbon waited far longer before writing a fix. With a harness that asked every five turns whether it already had enough to fix the bug, Carbon finished sooner, which the author reads as a behavioral habit rather than a knowledge gap.

Unconfirmed: The behavioral explanation is the author's inference from one modified sample trial.

Key facts
  • Astra named the cause within a minute and wrote a fix about three minutes later.
  • Carbon named the cause in about six minutes, then waited about 25 minutes to edit.
  • With check-ins, Carbon finished in 28 minutes instead of 37.
  • The post treats the shorter wait as evidence the delay is behavioral.
Explain this AI news with Vani AIWhat was announced, what is new and why it matters, in a new chat

From the original article

1 min read at kernel.dev

Minutes from the start of each run. GPT-6 Astra at max effort, Gemini 4 Carbon at high. In this sample trial, both models identified the cause early, Astra within a minute and Carbon within about six.

Hacker News · AI

By deenadz · kernel.dev

Read the full article at Hacker News · AI

A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.