What changed
On October 13, 2026, Google made Gemini 4 Argon broadly available to developers and enterprises globally via the paid Gemini API in AI Studio and its cloud platform, after its September 30 restricted Fairwind cybersecurity preview. Eligible Google AI Pro and Ultra users can access it in the Gemini app and AI Mode; individual Antigravity access requires Ultra. API introductory rates are $2 per million input tokens and $10 per million output tokens with a 95% cached-input discount. Google's October 13 footnote explicitly ends the introductory period November 30; rates rise to $4 input and $20 output per million from December 1. Google also advertises a 1M unified input-output token limit via a continuation protocol, which is not equivalent to an ordinary single response of one million tokens.
Why it matters
The immediate builder decision is whether Argon's coding, analysis and long-horizon capabilities justify trial workloads at the temporary lower rate, and whether an application remains economical when token rates double. A 1M-token continuation mode potentially changes how developers design long-running generation, but requires protocol-specific integration and realistic testing of latency, total tokens and failure recovery. Google's engineering and benchmark claims are useful leads, not guarantees for independent applications. Teams should benchmark representative tasks, record total token use and plan the December price step before committing to production.
Broad API access replaces the September restricted preview
Google first introduced Argon on September 30 to selected defensive-security testers. The October 13 release makes the public model available via the paid Gemini API to developers and enterprises globally. Google AI Pro/Ultra subscribers can use it in Gemini and AI Mode, while Google Antigravity access for individual subscribers is reserved for Ultra. The separately controlled Fairwind Program is expanding to hundreds of additional vetted defenders and is not a generally available unguarded API.
Introductory economics have a fixed end date
The official launch price is $2 per million input and $10 per million output tokens, with cached input priced at a 95% discount to the standard input rate. Google says the introduction expires November 30, 2026; from December 1 the published public-model price is $4/$20. At one million uncached input and one million output tokens, the model-only cost therefore rises from $12 to $24 before other charges, though real workloads vary in input/output balance and caching. Pricing should be rechecked before deployment.
The 1M limit uses a continuation protocol
Google advertises a one-million-token unified input-output limit through its Gemini API continuation protocol. This should not be presented as a promise that any ordinary single response can contain a million tokens. Developers need to review the protocol documentation and test how continuation, partial outputs, rate limits and failures behave in their own workloads.
Model quality claims depend on benchmark and configuration
Google cites leading Vals professional-workflow results and self-reported gains in engineering output, including more agentic code submissions and fewer rollbacks. Independent comparative coverage notes that Vals ranks Argon first on its index while Artificial Analysis places it behind some rival models on a broader index. Those assessments test different tasks and reasoning settings; Google's internal code-productivity figures and security claims are not independently replicated. As checked during this research pass, the linked external model-card PDF was still a coming-soon placeholder.