Gemini 4 Carbon is 5 times slower, coder says

Google's Gemini 4 Carbon is about 5 times slower than rival AI models at coding, YouTube creator Theo Browne said.

Browne posted the video on the 13th of November 2026. Google opened Carbon to paying developers on the 10th.

Nine rounds, one slow model

Browne showed logs from nine code review rounds. Four AI models got the same job at once.

Carbon finished four rounds in 16 minutes 44 seconds on average. Anthropic's Claude Opus 5.5 took 3 minutes 31 seconds.

In the other five rounds, his AI helper stopped waiting for Carbon.

An 11 second wait

Browne said Carbon waits more than 11 seconds before it starts to answer. On X, he wrote that this was 20 times longer than Anthropic's Claude Fable 5.2.

In one long job, Carbon read the same two files 16 and 17 times, his screen showed.

What Google said

Varun Mohan, who builds Google's Antigravity coding tool, replied on X. He wrote that the first reply is slow partly because reading a long request is costly.

A day later, Mohan wrote that Google was looking into how to speed it up.

The outside numbers

Artificial Analysis, an independent test site, says Carbon needs 31.3 minutes per task. The middle value for similar models is 13.1 minutes.

The site puts Claude Fable 5.2 at 45.3 minutes. Browne said such tests mislead because most are single questions.

Below we break down the wait

Carbon's 16 minutes 44 seconds is 1,004 seconds. That is 4.8 times the 211 seconds for Opus 5.5, by our arithmetic.

What is confirmed and what is not

The 5 times and 11 second figures come from Browne's own tests. Google has not shared its own speed figures.

Key facts

Theo Browne said Gemini 4 Carbon was about 5 times slower in his coding work.

He said Carbon waits over 11 seconds before it answers.

Google's Varun Mohan said the company is looking at ways to speed it up.