Google tests Gemini 4 Carbon checkpoint reportedly comparable to Claude Opus 5.5

Google is internally testing additional Gemini 4 checkpoints, with reports indicating that a variant named Carbon feels comparable to Claude Opus 5.5 for coding tasks. This development comes as Gemini 4 Argon appears in Antigravity selectors, signaling a potential imminent public release.

Internal Checkpoints and Comparisons

According to Business Insider, Google employees are evaluating multiple Gemini 4 variants. While Barium-B has been selected as the checkpoint for the public Argon release, Carbon represents a separate and potentially more capable iteration. Early feedback from employees testing Carbon through Jetski (Google’s internal name associated with Antigravity) highlights its strong coding performance relative to Anthropic’s current leader.

Antigravity Updates and Context Configurations

Recent changes to Antigravity 2.0 Insiders have introduced references to Gemini 4 Argon alongside three specific context configurations:

• 256K context by default.
• 512K context, consuming approximately 1.3x quota per turn.
• 900K context, consuming around 1.8x quota per turn.

These options mirror existing settings for Gemini Flash models, though the precise relationship between context size and reasoning effort remains unclear. Additionally, Gemini Web now supports low, medium, and high thinking effort settings across available models, potentially preparing the platform for the new release.

Rollout Timeline and Agent Features

Google officially announced Gemini 4 Argon on September 30, initially granting access to selected cybersecurity partners, with paid API customers and Google AI Ultra subscribers expected to follow. A wider rollout could begin as soon as next week, although no date has been confirmed. Concurrently, Antigravity has renamed its agent from Concierge to Chief of Staff, reinforcing Google’s focus on autonomous coding agents powered by increasingly capable Gemini models.