Gemini 4 Carbon early test results are in ahead of release

Open original

Google's next Gemini checkpoint, Carbon, is already circulating in early access and on LM Arena, with testers saying it beats the current Argon model at 3D work but still lags on frontend and UI and runs noticeably slower. The early outputs are based on arena matchups and hand-picked prompts, so they're signals, not benchmarks.

  • Testers report Carbon is clearly better than Argon at 3D tasks, like the voxel pagoda comparison, but still weak on UI and frontend output.
  • Speed is the main drawback: one tester says Carbon is roughly two to four times slower than Argon depending on the task.
  • The leak itself is awkward: Google's Trusted Tester Agreement asks testers to keep outputs private, yet Carbon samples keep showing up on X.

Google appears to be lining up Gemini 4 Carbon as the successor to Gemini 4 Argon, which has only been public for about a week. Business Insider first reported Carbon as a checkpoint inside Google's Jetski coding tool, and it now seems to be reaching outside testers. According to X user Lentils, a Carbon early access program has started, with partners getting the model and it being run on LM Arena. Lentils expects a release "very close" and says Carbon is noticeably better than Argon at 3D work, while frontend and UI output is still weak. It's also slower, which they put as "less lazy." Leaker Lyra backed this up by quoting Google's Gemini API Early Access Program docs, which describe an invite-only group getting upcoming models before launch. The catch is that the Trusted Tester Agreement says unreleased models, codenames, benchmarks, and outputs should stay private until launch. The steady stream of Carbon outputs on X runs directly against that. The early outputs come mostly from Lumina, who says Carbon has been on Arena since the day before and is a clear step up from Argon. The downside is speed: they put it at roughly two to four times slower depending on the task. Their voxel pagoda comparison, which pits Carbon against other frontier models, shows what they describe as noticeably higher quality than Argon. They also ran the SVG pelican on a bicycle test, where they again flagged weak UI and slow generation, but said Carbon otherwise "feels good" across the range of tasks they tried. The caveat matters: these are arena matchups and hand-picked prompts, not formal benchmarks, and nobody can be sure which checkpoint is actually being served behind a given answer. The stakes go beyond the chatbot itself. Logan Kilpatrick said Google processed around 25 trillion tokens in the first 72 hours after Argon launched, so demand is there and Google has a reason to keep shipping. Separately, broader Computer Use permissions have appeared in the Gemini desktop app, and Nano Banana 3 Pro reportedly sits on top of Gemini 4 with a November to December window in view. Competitors are moving quickly too: OpenAI's GPT-6.1 Terra has surfaced in Microsoft code, and Anthropic's Fable 5.5 is reaching partners. For Google, the question is whether Carbon fixes Argon's 3D problems without making users wait too long. The author plans to test it personally once it ships rather than trust the early impressions.

Ask about this article