Google is reportedly testing several versions of Gemini 4, with one internal model drawing comparisons to Anthropic’s Opus 5.5 on programming tasks. The reports suggest a fast-moving development effort, but they do not establish when Carbon might be released or what name it would use.
What the reports say about Carbon
Business Insider reviewed documents, screenshots and internal chats that describe three Gemini 4 variants: Argon, Barium and Carbon. Carbon was reportedly deployed on Google’s internal coding platform, Jetski, over the past few days.
According to the account, Carbon is said to outperform Argon mainly on programming tasks. One employee compared it with Anthropic’s strongest coding model, Opus 5.5, while cautioning that Carbon still needs more testing. That makes the comparison an early internal assessment, not a confirmed public benchmark.
Feedback about Argon has also varied by version and task. One employee said early Argon versions brought to mind the older Opus 5 on some coding work, though internal feedback was positive overall.
Argon’s role in Google’s model lineup
Earlier rumors had characterized Argon as a Flash model focused on speed rather than peak performance. Google’s description of its model roles points in a different direction: Argon is for frontier reasoning, Flash for speed and volume, Omni for generative media, and Gemma for lightweight, open-weights edge workloads.
In that framing, Argon is Google’s most capable reasoning model. The article compares its intended position with Anthropic’s Opus or OpenAI’s Astra. Carbon and Barium, meanwhile, may be checkpoints or updates within the Argon family rather than distinct model tiers.
There are clues that Carbon could eventually use the Argon name. One employee reportedly called it the “Gemini pro next model,” and internal documents show that the now-unveiled Argon was previously called “Barium-B.” But it remains unclear whether Carbon will arrive as an Argon update or as a standalone model.
Fast iteration, with release details still unknown
The pace of updates has prompted speculation that Google is getting better at using AI to help build new models. Google DeepMind employee Vedant Misra responded to the Business Insider report on X with, “Have you heard of recursive self improvement,” referring to the use of AI to build better AI. The source also says OpenAI and Anthropic have reported similar progress.
That idea offers one possible explanation for rapid model iteration, but the reported development timeline alone does not show how much AI contributed to creating Carbon. The available account focuses on internal reports and employee comments; it does not provide a public evaluation demonstrating Carbon’s coding performance.
Google has not announced a Gemini 4 launch date. Still, the company appears to be preparing its products: the Gemini app now shows an “Automatic” mode for 3-series models and adjustable reasoning intensity from low to high. Google AI Studio has surfaced an “Ultra” mode described as offering “advanced skills and tools.” Google’s Logan Kilpatrick also confirmed that the team is working to get the most out of Argon.
What to watch for next
The key open questions are whether Carbon becomes part of Argon, whether it will be offered as a separate model, and how its coding ability holds up after more testing. Until Google provides more details, the comparisons should be treated as reported internal impressions rather than a final measure of performance.
- Model names: Argon, Barium and Carbon are reported Gemini 4 variants.
- Reported strength: Carbon is said to improve on Argon mainly in programming tasks.
- Release status: Google has not announced a Gemini 4 launch date, and Carbon’s eventual naming remains uncertain.