News

Releases & Models

CONFIRMED: Google opens Gemini 4 Argon to everyone and rejoins the frontier race

After delays and turmoil at DeepMind, Argon leaves restricted access and reaches the API, app and Search, priced at US$2 and US$10 per million tokens until November 30, with a one million token window.

Ilustração editorial de uma estrela luminosa em degradê azul e violeta, no estilo do Gemini, irradiando luz sobre um globo terrestre conectado, simbolizando o lançamento global do Gemini 4 Argon pelo Google

Google made Gemini 4 Argon generally available worldwide on October 13, two weeks after announcing it with access limited to cybersecurity defenders. The model now reaches developers, enterprises and consumers through the Gemini API, Google's application programming interface, as well as Google AI Studio, Google Antigravity, the Gemini app and AI Mode in Search. A tightly controlled promise has become a product any team can test.

A comeback after turbulent months

The launch carries more weight than its spec sheet. According to Reuters, it arrives after months of delays and a leadership overhaul at DeepMind, Google's artificial intelligence (AI) lab. In September, the company dropped Gemini 3.5 Pro, which chief executive Sundar Pichai had promised for June. DeepMind cofounder and chief executive Demis Hassabis stepped aside, and several leaders of the Gemini team left the company. The agency reads Argon as Google's attempt to regain ground against OpenAI and Anthropic.

Aggressive pricing, with an expiry date

On the API, introductory pricing is US$2 per million input tokens and US$10 per million output tokens, with a 95% discount on cached input. Tokens are the chunks of text the model processes and bills for. The rate lasts until November 30, 2026; after that it doubles to US$4 and US$20. For anyone building automation or products, long instructions and repeated document bases become far cheaper with caching, but any cost projection must account for full pricing from December onward.

One million tokens in a single window

The context window is unified at 1,048,576 tokens, shared between input and output. Default synchronous output stops at 65,536 tokens, but a continuation protocol allows longer generation within the same million token budget. The model accepts text, images, video, audio and PDF files, the portable document format, and answers only in text: it does not generate images or audio and does not work with the Live API, Google's interface for real time conversations. Anyone who needs live voice still depends on other models.

Built for long work

Google positions Argon for complex, multistep tasks in coding, science, law and cybersecurity, what the company calls deep work. According to Google, it leads the Vals Index, which measures performance in sectors such as finance, coding, law and tax. Teams using Argon in Antigravity increased agentic code submissions by more than 35% and saw nearly a third fewer rollbacks. The Chrome team separated core webpage handling from a legacy folder inside Chromium's 40 million line codebase, a refactor across hundreds of files that would have taken months. The ADK team, behind Google's Agent Development Kit, says the model optimized its Python code overnight, cutting in memory preparation of long sessions by 40% and speeding up cold imports by 21%.

Who gets what

In the Gemini app and AI Mode, Argon reaches Google AI Pro and AI Ultra subscribers; Antigravity with Argon is limited to Ultra. The app gains a toggle between chat and tasks, a mode previously called Spark, which runs background work across Google apps and more than 40 integrations such as Instacart, OpenTable, Adobe and Shopify. Google promises six months of free access to Antigravity's highest usage tier for 10,000 scientists, with gradual onboarding and selection, and is expanding to hundreds of new partners the Fairwind program, which gives vetted defenders the model without standard cybersecurity guardrails. Deloitte, LexisNexis, Ambience Healthcare and Isomorphic Labs tested the model early.

The MaxAssistant view

The real news is access: Argon is no longer a showcase for a few and has become a genuine production option, cheap for a limited time. The productivity figures come from Google itself and still need independent evaluation. For anyone building products or automation, the sensible path is to use the window until November 30 to test a long, bounded workflow, measure retries and final cost, and only then decide whether Argon joins the stack.

Sources

Google: Gemini 4 Argon: Built to solve your hardest problems; Reuters: Google launches flagship AI model Gemini 4 Argon for developers, businesses; 9to5Google: Gemini app rolling out Gemini 4 Argon to AI Pro and AI Ultra; Blockchain.News: Google Launches Gemini 4 Argon AI Model for Enterprise and Developer Use