Google's Gemini 4 Carbon reportedly matches Anthropic Opus 5.5 coding

By Splitfeed EditorialLast updated Oct 10, 2026, 10:00 UTC

On October 10, 2025, The Decoder reported, citing Business Insider, that a Google employee had compared the coding performance of an unreleased internal model called Gemini 4 Carbon to Anthropic's Opus 5.5, suggesting the two are roughly matched. The claim surfaced even as Gemini 4 Argon, an earlier variant in the same generation, had not yet reached wide availability. Separately, new modes appearing in the Gemini app and Google AI Studio have led observers to interpret Google's product pipeline as moving toward a broader Gemini 4 rollout, though neither Google nor Anthropic has publicly confirmed any of the performance figures.

The context that gives this story weight is the competitive positioning it implies. Anthropic's Opus line sits near the top of most public coding benchmarks, and a claim that Google has an unreleased model already matching it, before the previous variant has even shipped broadly, would suggest Google is moving faster internally than its public release cadence indicates. What is genuinely in dispute is whether a single employee comparison, made without published benchmark methodology or third-party verification, carries any meaningful signal at all. Internal comparisons can reflect cherry-picked tasks, different prompt conditions, or simply informal impressions rather than rigorous evaluation.

No reactions from the Pro-AI, Anti-AI, or Middle Ground camps have been published in response to this story yet. The Pro-AI camp would typically treat such a report as evidence of accelerating competition driving rapid capability gains, framing it as a sign that frontier coding performance is becoming more widely distributed. The Anti-AI camp would likely focus on the unverified nature of the claim, arguing that internal employee comparisons are a form of hype that outpaces reproducible evidence. A Middle Ground position would probably acknowledge the plausibility of rapid iteration while insisting that external benchmarks and independent evaluations are the only reliable basis for drawing conclusions about relative model performance.

The clearest resolution would come from a formal Gemini 4 Carbon release accompanied by published benchmark results on standard coding evaluations, or from independent third-party testing that pits Carbon against Opus 5.5 under controlled conditions. A broader Gemini 4 launch announcement from Google, which the new app modes may be foreshadowing, would be the next concrete event to watch.