Gemini 4 Argon: Pricing and Availability

Google has released Gemini 4 Argon, which it calls its latest frontier model, and it is available globally from October 13. Google says it is aimed at long, multi-step work in coding, finance, legal, cybersecurity and science. It is available through the API in AI Studio and the Gemini platform, in Google Antigravity for Google AI Ultra subscribers, and in the Gemini app and AI Mode in Search for Google AI Ultra and Pro subscribers. The launch price is $2 per million input tokens and $10 per million output tokens. This guide breaks down what Google claims, which claims are checkable, and how to decide whether to switch. For a short reference to this page, here is our Gemini 4 Argon launch guide. All statements come from Google's own announcement post; benchmark and productivity figures are Google's, not independent results.

What Google announced

Argon is the first model announced from the Gemini 4 family, and Google says more members are coming. Google published an earlier post about Argon on September 30, and the global rollout began on October 13, alongside a developer post and a safety post. The company says Argon works longer and harder on complex scientific, coding and agentic problems, helped by what it calls an industry-leading 1M unified input-output token limit via the Gemini API and a new continuation protocol that developers should read about before setting up.

On benchmarks, Google says Argon outperforms other frontier models across several domains and ranks first on the Vals Index, which it describes as measuring model performance across economic sectors like finance, coding, legal and tax work. Google also says it leads on the Gray Swan Indirect Prompt Injection benchmark, cutting attack success across 15 adaptive attempts to 0.7 percent, and scores 87 percent on the Embodied Reasoning benchmark. Those are Google's own reports, and the page does not give full methodology, so independent evaluations will matter.

Pricing and where to get it

Google calls $2 per million input tokens and $10 per million output tokens an introductory price, with cached input tokens at 95 percent off. The word introductory suggests it could change, so if you are budgeting, check the pricing page when you plan a rollout rather than relying on this number. The ways to access Argon are:

• Developers: the paid tier of the Gemini API in AI Studio and Android Studio, and AI Studio and Antigravity for Google AI Ultra subscribers.
• Enterprises: Gemini for business, the API and Antigravity.
• Everyday users: the Gemini app and AI Mode in Search, for Google AI Ultra or Pro subscribers.
• Scientists and researchers: Google is offering 10,000 scientists the highest usage tier in Antigravity free of charge for six months, onboarded on a rolling basis and subject to eligibility, capacity and selection.
• Cybersecurity defenders: they can apply to the Fairwind Program for access to Argon without cyber guardrails on a rolling basis, and Google says it is starting to expand to hundreds of additional partners.

Notably, the free Gemini tier is not on that list. If you do not subscribe to Google AI Pro or Ultra, you cannot use Argon in the consumer app today, according to the post.

The coding claims, in Google's words

Much of the post is about engineering work at Google. Google says Argon in Antigravity helped teams increase agentic code submissions by more than 35 percent while reducing code rollbacks by nearly a third. It gives several examples. In Chrome, Argon helped separate core webpage handling from a legacy folder of about six million lines inside Chromium's 40-million-line codebase, finishing an architectural refactor across hundreds of files in weeks rather than months. In Google's open-source Agent Development Kit, Argon is said to have cut in-memory session preparation time by 40 percent in long sessions and sped up cold imports by 21 percent. Google also reports that it made its open Gemma models nearly three times as fast, a 2.85x end-to-end inference speedup on an A100 through vLLM with token parity.

These are internal case studies with Google's own teams, so they show what the company believes the model can do under favorable conditions with its own tooling, such as skill rulebooks and multi-agent loops. Your results will depend on your codebase, tests and how much supervision you build in.

Business and science claims

Google says Argon combines reasoning, visual perception, domain knowledge and stronger writing, and that it follows style instructions better than previous Gemini models and answers more directly with less hedging. It describes use in drug discovery workflows at Isomorphic Labs, where Argon synthesizes complex PDFs and patents in hours rather than weeks, and security work with Wiz, where it helped find and fix serious flaws affecting government and healthcare systems. The post also lists science results in quantum computing, fusion, rare disease analysis and enzyme design, including a claim that an Argon-written implementation of a quantum subroutine cut spacetime cost by more than 100x compared with a published baseline. These are headline results chosen by Google, and some depend on private evaluations, such as those cited from Gamow Labs.

Safety claims

Google says Argon is its most comprehensive safety and security stack yet, evaluated under its Frontier Safety Framework with third-party partners. It reports that external bio red teaming could unlock less than 1 percent of tested bio sub-tasks, that violations on an adversarial set fell nearly two-fold versus previous models, and that shortcut attempts in coding benchmarks dropped to 2 percent. It also describes an automated misalignment monitor that reads an agent's reasoning and actions to block unintended workarounds. A model card is available for those who want details. As always, these are the developer's own figures, and a safety stack reduces risk but does not remove it, particularly when agents have access to your files, email or systems.

Who is this for

• Engineering teams running large refactors, performance work or code review who want to test a long-context, agentic model.
• Legal, finance and research teams that need document-heavy analysis and are prepared to verify output.
• Security teams that qualify for the Fairwind Program.
• Scientists and students in research who may qualify for the six-month Antigravity offer.
• Casual users who just want a chat assistant probably do not need to pay for a higher subscription tier on the strength of this announcement alone.

What to check before switching

• Cost on your workload. A price per million tokens matters less than total tokens for your tasks. Long-running agentic work can use many output tokens, so run a pilot and measure.
• Access tier. Consumer access requires Google AI Pro or Ultra, and Antigravity needs Ultra.
• Introductory pricing. Confirm what the price becomes later.
• Context and continuation. The 1M unified token limit and the new continuation protocol change how long jobs are structured, so read the developer docs.
• Evaluate on your own data. Build a small test set from real tasks and compare against your current model, rather than trusting a leaderboard.
• Guardrails for agents. Decide which tools and data an agent can touch, and keep human approval for irreversible steps.

Weak points and open questions

Because the post is a company announcement, most performance evidence comes from Google itself. Benchmark names are given but not full tables, the Vals Index claim is a ranking statement without displayed scores, and many case studies cannot be independently reproduced. The announcement also says nothing about how Argon compares on speed or cost with smaller models in the same family, which will matter for routine tasks. Independent evaluations and developer feedback over the coming weeks will show how much of this carries over to everyday work.

Bottom line

Gemini 4 Argon is a serious frontier release aimed at long, agentic work, with a clear price, a wide set of access routes and unusually detailed safety claims. If you build software or run document-heavy workflows, a pilot is worthwhile now. If you are a general chatbot user, there is little pressure to move before reviews and independent tests come in.