Google DeepMind has unveiled Gemini 4 Argon, its latest frontier model, with one of the most aggressive spec sheets of the year: a one-million-token output limit, up from 64,000 in the previous generation, and introductory pricing pitched squarely at cost-sensitive builders. Announced September 30, 2026, Argon is aimed at long-horizon work in software engineering, enterprise knowledge work, and cybersecurity defense, the kind of tasks where a model must reason across enormous codebases and document sets before answering.
On benchmarks, Google claims Argon leads or ties on 13 of 18 disclosed tests against OpenAI's GPT-6 Astra and Anthropic's Claude Opus 5.5. The highlights include 77.9 percent on DeepSWE v1.1, a test of real-world software engineering, ahead of Opus 5.5 at 74.2 percent; 19.6 percent on Harvey's Legal Agent Benchmark against 5.4 percent for Astra; and 51.3 percent on AutomationBench versus 42.5 percent for Opus 5.5. Google also cites 68.9 percent on the Vals Index and strong scores on its internal cybersecurity benchmarks, and Artificial Analysis has independently scored Argon 53 on its Intelligence Index.
The record is not unbroken. GPT-6 Astra still leads by roughly ten points on FrontierSWE v2, and Opus 5.5 holds a clear advantage on Terminal-Bench 4.0. Every score so far comes from Google's own testing or from a gated early-access cohort, so the numbers are claims to be confirmed rather than settled measurements. Argon's strengths cluster in coding and agentic tasks; its weaknesses sit elsewhere.
Pricing is where Google is clearly hunting. The introductory rate is two dollars per million input tokens and ten dollars per million output tokens, with cached input discounted by 95 percent. Google says the rates will later rise to four and twenty dollars, so builders should treat the opening price as a trial balloon rather than a promise.
Then there is the matter of who can actually use it. Argon is rolling out first to trusted cyber defenders through Google's Fairwind Program and to U.S. government pre-release programs, with paid API customers and Google AI Ultra subscribers next and no date announced. In other words, one of the best-measured models Google has ever shipped is currently available to almost no one. Google says thousands of its own employees are already using Argon for coding, research, and infrastructure optimization.
Wall Street noticed. Alphabet shares rose on the announcement, and JPMorgan maintained its Overweight rating while warning that Google still needs to re-establish itself at the AI frontier for both the industry and investors. Argon's message is that Google intends to compete on cost per completed task and sheer output capacity rather than raw capability alone. Whether that is enough depends on when the rest of us actually get to try it.
