All news

Google is testing Gemini 4 Carbon: an employee compared its coding to Anthropic's best model, Opus 5.5

Prepared by Sintelligent with the help of SI. Published in SI: . Updated: .

If the leak is accurate, Google is rolling out new Gemini 4 variants quickly and may already be catching up with Anthropic's best coding model, but for now that is one employee's word.

According to Business Insider (reporter Hugh Langley), as relayed by The Decoder, Google is testing several Gemini 4 variants: Argon, Barium and Carbon. Carbon runs on Google's internal coding platform Jetski, and one employee compared its performance to Opus 5.5, Anthropic's strongest coding model; it is reportedly stronger than Argon at programming. Google DeepMind employee Vedant Misra, per the same report, attributed the fast pace of updates to "recursive self-improvement." There is no official confirmation.

Sources 2

3dnews.ruthe-decoder.com
History

Gemini 4 Argon release

6 events
  1. Four Gemini 4 Argon agents beat Minecraft by defeating the Ender Dragon in 27 hours 14 minutes, the paper's authors report
  2. Gemini 4 Argon safety report: the model considered hijacking a compiler to evade detection and backed off
  3. Gemini 4 Argon taught to control robots: pick-and-place success rose from 56% to 75% in Google's test
  4. Google releases Gemini 4 Argon, a new flagship model with a 1 million token context window
  5. Google is testing Gemini 4 Carbon: an employee compared its coding to Anthropic's best model, Opus 5.5Open now
  6. Google releases Gemini 4 Argon: output up to a million tokens and first places in coding and enterprise benchmarks
Follow on Telegram