Google has introduced two frontier AI models designed for demanding enterprise workloads at competitive prices, renewing its position in the AI race.
Gemini 4 Argon became broadly available October 13 after an earlier limited release to cyber defenders. Gemini 4 Carbon was announced October 28 with restricted access. Both models target complex coding, cybersecurity, and agentic workflows.
While OpenAI, Anthropic, and Meta captured headlines throughout 2026, Google maintained a lower profile.
How do the models compare? On Artificial Analysis’s Intelligence Index v5, Argon scores 50, while Carbon scores 59, the index’s highest score at the time of writing. Carbon does not lead across every coding and knowledge-work evaluation, however.
In a Coding Agent Index comparison, Carbon with Antigravity CLI scores 59 versus 62 for Claude Opus 5.5 with Claude Code. These rankings measure specific criteria, not which model is best across workloads.
Google says Argon can help developers debug, optimize, and modernize large systems. In one internal example, Argon agents optimized a video decoder to run 2.7 times faster than an existing Rust implementation, while producing identical results. That’s compelling for enterprises with complex codebases, though results should be evaluated in your own environment.
Cybersecurity is another strength. Argon can identify and help remediate software vulnerabilities. Google said Argon helped cybersecurity firm Wiz uncover a serious vulnerability in healthcare software that earlier frontier models missed. Argon also performs well on legal, financial, and other knowledge-intensive tasks, including multimodal and large-context analysis, according to Google.
Carbon shows further gains in vulnerability detection, exploitation testing, and patching, which matter to enterprises with sensitive data and critical systems. Google reported scores of 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for testing exploitability, and 71% on CWE-bench for patching. These benchmarks don’t guarantee real-world security outcomes.
Both models were announced with introductory API rates of $2 per million input tokens and $10 per million output tokens. Argon’s discount ends November 30, rising to $4 and $20 on December 1. Google hasn’t specified when Carbon’s discount ends.
But token fees are only part of the cost equation. Integration effort, accuracy, human review, workflow changes, and the work it takes to embed AI into existing processes add to the total cost of ownership.
Access is another consideration. Argon is available through the Gemini API and Google AI Studio. Google AI Pro and Ultra subscribers can access Argon through the Gemini app and AI Mode in Search. Ultra subscribers have access through Google Antigravity.
Carbon remains limited to selected cyber defenders and trusted partners while Google continues to test its safeguards.
Stronger agents require stronger governance: controls over data access and permitted actions, monitoring for unexpected agent behavior, and defined points for human intervention. Organizations should weigh how a model fits into existing processes rather than treating new models as a solution on their own.
Google now offers credible alternatives to models from Anthropic, OpenAI, and Meta. But benchmark leadership alone isn’t a reason to switch.
Businesses should identify workflows where stronger reasoning, better coding, or cybersecurity capabilities could deliver measurable results. Test Argon against your current stack, comparing accuracy, task completion, speed, and total cost, including integration and human oversight. Most businesses will need to wait for broader Carbon access before running their own evaluations.
Google has reestablished itself as a serious contender for enterprise AI workloads. Its lasting impact will depend on how these models perform in real business environments.