Vibeleaderboard
← All Intel
Intel / post

Gemini 4 Carbon day-zero review: strong coder, weak judgment

Source
x.com
Date
Dan Shipper@danshipper

BREAKING: @Google just dropped Gemini 4 Carbon. It’s a much better coder than Argon—but I’m probably keeping Argon for writing. We’ve been testing it at @every across coding, writing, design, and knowledge work. It can work for hours on an ambitious project, but it doesn’t always know when to stop. Here’s your day-zero vibe check: • Coding: On one run of our Senior Engineer benchmark, Carbon scored 44%, versus Argon’s 21%. A big improvement, though still middle of the pack. In a separate test, @hammer_mt gave it a backlog to work through. It identified 6 tasks as ready, then did 15—including work someone else was already doing. He stopped it after almost 4.5 hours. • Big builds: @RandyCounsman gave it 3D scans of our podcast studio and an equipment list. In one attempt, it built a browser-based model accurate enough to check whether new gear would fit without getting into the shot. He estimates Carbon also wrote about 95% of a big update to Everycut, our internal video editor. • Code quality: @kieranklaassen values its tests and documentation on large projects, including its record of approaches it considered and rejected. Fable remains his first choice, but Carbon has…

Read the full post on X
Why it matters

Carbon scored 44% versus Argon's 21% on one Every coding run, but it overran scope on a backlog, doing 15 tasks when 6 were ready. Scope it to one big job with a clear finish line and review the output.

Terms in this piece · Glossary
  • benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.
  • token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
More from Dan Shipper
Comments

Checking sign-in…

Loading comments…