Gemini 4 Argon Opens Up: Paid API, AI Pro and Ultra Get Google's Frontier Model — Intro Pricing Ends Nov. 30

This is an analysis of public material: Google's official posts and developer documentation plus press coverage. We did not run the model or reproduce any benchmark.

Who should care

• Teams building on the Gemini API. A new top-end model is live on the paid tier today, and the discount window is roughly seven weeks. Budgets written now need to assume December's list price.
• Google AI Pro ($19.99/month in the U.S., per press reports) and Ultra subscribers. Argon shows up as a model choice in the Gemini app and AI Mode at no extra charge, with Ultra getting higher limits.
• Security teams. Google says it is expanding its Fairwind Program, which offers Argon without cyber guardrails to vetted defenders, to hundreds more partners.
• Free users. Not included in this release, in either the consumer app or the API.

What Google confirmed

In the Oct. 13 launch post, Google DeepMind's Koray Kavukcuoglu said Argon is "globally available today for developers and enterprises via the API." A separate Gemini app post said Pro and Ultra subscribers can now use it.

The Gemini API pricing page adds detail: cached input is $0.10 per million tokens (rising to $0.20), with implicit caching only and no explicit cache support. Batch requests cost half the standard rate. The page says Argon is not available on the free tier. Google's developer post also lists AI Studio and Android Studio as entry points.

The Gemini app also gained a "tasks" mode, switched on with a toggle in the chat window. Google says Gemini can keep working while you are away from your computer, and it now connects to more than 40 apps, including Instacart, OpenTable, Adobe and Shopify. Antigravity's agentic coding interface is available inside the Gemini app for macOS.

For researchers, Google is offering up to 10,000 scientists six months of free access to Antigravity's highest usage tier, subject to eligibility, capacity and selection.

The performance claims, and who made them

Google reports a No. 1 ranking on the Vals Index (finance, coding, legal and tax tasks), 87% on the ERQA embodied-reasoning benchmark, and an indirect prompt-injection attack success rate of 0.7% across 15 adaptive attempts on Gray Swan's benchmark. Its Sept. 30 announcement put Argon at 77.9% on DeepSWE v1.1, a long-horizon software engineering test. SiliconANGLE noted that Google's chart shows Claude Opus 5.5 at 74.2% on that test, and that the comparison comes from Google itself.

The one outside figure in early coverage comes via Engadget, which reported that Artificial Analysis scores Argon level with OpenAI's GPT-6 Astra on its Intelligence Index at about 60% of the cost per task at introductory prices. We have not checked that figure against Artificial Analysis directly, and the cost advantage shrinks once the intro pricing ends.

Still unconfirmed

• Country and language coverage. Google calls the API release global but publishes no country list for the app or AI Mode. Search Engine Journal reported that the AI Mode rollout is gradual, that its own Pro account did not show Argon at first, and that no language was specified.
• Tasks mode regions. SiliconANGLE, citing Google's plans page, said tasks mode is limited to select countries for now. Which countries is unconfirmed.
• Exact usage limits. Google says Ultra gets higher Argon limits than Pro but has not published numbers in these posts.
• Rest of the family. Google said it looks forward to "bringing you the rest of the Gemini 4 family" but gave no dates.

What it means for your bill

The headline change for developers is a built-in price cliff. Every cost model built on today's $2/$10 rates will be off by a factor of two on Dec. 1. Because output pricing includes thinking tokens, workloads that lean on long reasoning can produce output bills well above what the visible response length suggests.

The 1M-token unified limit makes single-pass work over large codebases or document sets practical, but filling that window at list price costs about $4 in input alone per full-context call.

Limitations

This piece relies on Google's Oct. 13 posts and documentation and same-day press coverage. Benchmark and safety figures are Google's, with limited independent replication so far. Regional availability and per-tier limits were not published in the sources consulted, and a staged rollout means access can change over the coming days.

Last verified 2026-10-14.