rkj dev

Google Tests Gemini 4 Carbon Model on Internal Jetski Coding Tool

Engineers at Google are experimenting with a new Gemini 4 checkpoint called Carbon, drawing early internal comparisons to competing coding models.

Illustration of software engineers working at computer monitors in a tech development lab
Illustration: Engineers analyzing internal model outputs.AI-generated illustration

Key takeaways

  • Google staff are internally evaluating a new Gemini 4 checkpoint codenamed Carbon on the company's Jetski coding platform.
  • Internal testing follows the late-September 2026 announcement of the Gemini 4 family, which includes the upcoming Argon release.
  • One employee cited by Business Insider compared Carbon's coding feel to Anthropic's Opus 5.5, while noting further testing is required.
  • Google's internal development line included checkpoints codenamed Argon, Barium, and Carbon, with Barium-B selected as the public Argon model.
  • Google said it would first make the model available to partners in its Fairwind Program before a wider rollout.

Google staff have begun evaluating a new artificial intelligence checkpoint codenamed Carbon inside the company's internal development environments. The deployment, reported by Business Insider, gives engineers access to an updated Gemini 4 variant through Google's internal coding workbench, known as Jetski, just as the company prepares to publicly release its Gemini 4 Argon model.

Carbon represents the latest iteration in a sequence of internal builds developed for Google's next-generation Gemini 4 series, which Google announced in late September 2026. According to internal documents reviewed by Business Insider, the system has sparked active discussion across internal messaging channels.

Illustration of a programmer reviewing complex code and system metrics on dual screens
Illustration: Internal software testing on dedicated developer platforms.AI-generated illustration

Internal Testing on the Jetski Development Platform

The arrival of Carbon on Jetski points to a fast iteration cycle inside Google's AI teams. As detailed by Business Insider, Google's internal development cadence has produced multiple builds under distinct chemical-element labels, including Argon, Barium, and Carbon.

Internal codenames at Google do not always map directly to consumer product names. Internal records indicate that a checkpoint labeled "Barium-B" was designated as the release that will ship publicly under the name Gemini 4 Argon. Whether Carbon will eventually be delivered to the public as an incremental update to Argon or launch under another name in the Gemini 4 model family has not been determined.

Google declined to comment on the internal testing or specific roadmap decisions, according to Business Insider. The testing appears to mirror Google's historical pre-release patterns, where internal dogfooding has historically preceded public rollout by weeks, according to market analysis from Newsquawk.

How Staff Compare Carbon to Competing Coding Models

Coding capabilities have become a core battleground among frontier model providers. Early feedback on Carbon suggests noticeable progress over earlier internal checkpoints. One Google employee speaking to Business Insider remarked that Carbon "feels like Opus 5.5" for coding, referencing Anthropic's model tailored for long-horizon agentic programming. The same employee cautioned that more comprehensive testing is still needed to establish its sustained performance.

Other internal chat messages reviewed by the publication showed employees welcoming the upgrade. One employee posted that "Carbon is really good!" while another observed that the "new Gemini pro next model feels kinda nice," using the internal "Next" label Google often applies to incoming model revisions.

Illustration of two developers reviewing performance graphs together
Illustration: Technical teams comparing pre-release benchmark feedback.AI-generated illustration

The positive internal reaction marks a shift from early feedback on Argon. While early internal versions of Argon were generally well received, at least one engineer told Business Insider that early Argon builds felt behind on certain programming duties, comparing its behavior to Anthropic's older Claude Opus 5 release. The latest updates might close that gap, but Google is not guaranteed to release them.

Argon Rollout and the Fairwind Security Program

While Carbon remains confined to internal staff on Jetski, Google's public rollout plans for Gemini 4 Argon remain underway. When announcing Gemini 4 in late September 2026, Google reported that Argon delivered frontier-tier results across specialized domains, highlighting benchmarks in legal and financial work as well as programming tasks.

Google also highlighted defensive cybersecurity as a priority for Argon. In line with that focus, Google stated that the model will first be made accessible to participants in its Fairwind Program—an initiative dedicated to identifying cyber vulnerabilities and stress-testing defensive systems—before opening up to broader public availability.

Beyond standalone model builds, Google announced an enterprise-oriented "universal" Gemini AI agent designed for workplace automation during the week of October 5, 2026, as documented by Business Insider. That release illustrates Google's parallel efforts to deliver finished productivity tools while refining the frontier models that power them.

Illustration of cybersecurity analysts monitoring defense infrastructure dashboards
Illustration: Security testing workflows ahead of a public software rollout.AI-generated illustration

Why the Internal Checkpoints Matter for Google

Google faces sustained pressure from rivals including OpenAI and Anthropic, both of which have found strong traction among software engineers using agentic coding products. As noted by Newsquawk, pre-release reporting on model checkpoints reflects broader market scrutiny over whether major research labs can maintain leadership across technical benchmarks.

For enterprise customers and developers, internal testing of Carbon indicates that Google is treating code generation and agentic task execution as critical product priorities. However, as AI Weekly emphasized, Carbon has no public release date, no public benchmarks, and has not been independently evaluated by outside engineers. Whether the improvements reported by Google staff materialize in the public Argon release—or in a later Gemini 4 release—depends on which model weights Google ultimately chooses to deploy.

Frequently asked questions

What is Gemini 4 Carbon?

Gemini 4 Carbon is an internal model checkpoint being tested by Google staff on the company's internal coding platform, Jetski. It follows previous internal versions codenamed Argon and Barium.

How does Carbon compare to other AI models?

One Google employee quoted by Business Insider stated that Carbon feels comparable to Anthropic's Opus 5.5 for coding tasks, though they added that further testing is required. No public benchmarks for Carbon have been released.

When will Gemini 4 Argon and Carbon be available to the public?

Google has not announced a public release date for Carbon, and it is uncertain whether it will be released publicly. Google stated that Gemini 4 Argon will first be offered to partners in its Fairwind Program for cybersecurity testing before a broader public rollout.

Sources

  1. Google employees praise latest Gemini 4 internal modelBusiness Insider · Oct 9, 2026
  2. Google Staff Test New Gemini 4 'Carbon' Build on JetskiAI Weekly · Oct 9, 2026
  3. Google (GOOGL) is preparing to publicly roll out its new Gemini 4 model, internally known as Argon, reports…newsquawk.com · Oct 9, 2026

How this story was made: the newsroom picked it up from Techmeme and Google News, gathered the full text of the sources above, and drafted it with AI assistance. Every factual claim was then checked against those sources before publishing (19 claims checked). Illustrations marked as AI-generated are not photographs. Spotted an error? Tell us.

#Google #Gemini 4 #Gemini 4 Carbon #Gemini 4 Argon #AI coding #Anthropic

Published October 10, 2026 at 01:43 UTC