Google's Gemini 4 Argon Leads AI Benchmarks in Professional Tasks
Google has unveiled Gemini 4 Argon, its most advanced AI model yet, which leads or matches top competitors in 14 out of 19 benchmarks. The model excels in professional knowledge work, outperforming GPT-6 Astra and Claude Opus 5.5 in tasks like financial research, legal drafting, and business automation. While coding results are mixed, Argon sets a new high on DeepSWE v1.1, a long-horizon software-engineering suite, though it trails in certain terminal and systems-level tests. Notably, the model can generate responses up to one million tokens, enabling extended reasoning and document handling.
Access to Gemini 4 Argon is currently restricted to trusted testers and cybersecurity teams, with paid API customers and Google AI Ultra subscribers expected to follow. Early results highlight Argon's strengths in professional workflows, such as business automation and legal research, where it outperforms rivals by significant margins. For instance, it scores 51.3% on AutomationBench and 65.4% on Vals Finance Agent v2, demonstrating its ability to handle complex, multi-step tasks.
In coding, Argon's performance is more nuanced. While it leads on DeepSWE v1.1 with 77.9%, it lags in terminal-bench-style evaluations and certain infrastructure tasks. Google has also prioritized cybersecurity, granting early access to defenders who identified a critical vulnerability in healthcare software that earlier systems missed. Google CEO Sundar Pichai emphasized a responsible rollout, stating, 'Importantly Argon has frontier safeguards and we are rolling it out responsibly.'
The controlled release strategy allows for observation of potential failure modes before broader deployment. As the model moves into paid hands, its success will be measured by real-world integration, including latency, tool reliability, and cost-effectiveness. While benchmark leadership is provisional, Argon's strengths in professional knowledge work signal a shift toward measurable economic utility in AI development.