Stories
Google Employees Test New Gemini 4 Model Carbon Amid AI Coding Race
- •Google is internally testing a new Gemini 4 artificial intelligence model codenamed Carbon to enhance its coding capabilities.
- •Employees who tested Carbon report its performance is comparable to Anthropic's Opus 5.5, though further testing is required.
- •Internal documents show Google has been evaluating multiple Gemini 4 versions under the codenames Argon, Barium, and Carbon.
- •The testing reflects Google's ongoing efforts to compete with OpenAI and Anthropic in AI-assisted software development and complex engineering tasks.
- •Google recently announced a universal Gemini AI agent for workplace use that can integrate with numerous platforms and work toward specified goals.
Key Questions Answered
Carbon is an internal version of Google's Gemini 4 artificial intelligence model that is currently being tested by Google employees on the company's Jetski coding platform. Employees reviewing the model have noted that its coding performance feels comparable to Anthropic's Opus 5.5.
Internal documents reviewed by Business Insider reveal that Google has been evaluating several versions of Gemini 4 under the codenames Argon, Barium, and Carbon. One of these documents specifically identified Barium-B as the version chosen for release under the Argon name.
The new universal Gemini AI agent can integrate with various workplace platforms including Google Workspace, Microsoft 365, Slack, Jira, Confluence, Git, BigQuery, Databricks, Postgres, and Snowflake. Additionally, it can connect to Model Context Protocol (MCP) servers operating either within or outside an organization's network.
