Gemini 4 Argon: What It Is, How to Access It, and How It Compares

Gemini 4 Argon is Google DeepMind's new frontier AI model, built for long, complex work in coding, finance, legal, science and cybersecurity. Announced on September 30, 2026 for a small group of security partners, it became widely available on October 13, 2026 in the Gemini app, AI Mode in Search and the Gemini API. API pricing starts at $2 per million input tokens and $10 per million output tokens.

What is Gemini 4 Argon?

Gemini 4 Argon is the first model in Google's Gemini 4 family and the company's new flagship. In its launch post, Google DeepMind's Koray Kavukcuoglu positions Argon for deep, multi-step work rather than quick chat: large code migrations, enterprise document analysis, research and security auditing.

Argon is also the model Google built instead of the cancelled Gemini 3.5 Pro. It is the first new Gemini generation since Gemini 3 in November 2025, according to an explainer from andrew.ooo.

Google says Argon writes more directly than earlier Gemini models, with less hedging, less repetition and fewer clichés, and keeps users updated more often during long tasks.

When was Gemini 4 Argon released?

Argon rolled out in two stages:

• September 30, 2026: Google announced Argon and gave access only to vetted cyber defenders in its Fairwind Program.
• October 13, 2026: Google opened Argon globally to developers through the paid Gemini API, to Google AI Ultra subscribers in Google Antigravity, and to AI Pro and Ultra subscribers in the Gemini app and AI Mode in Search.

How do you get access to Gemini 4 Argon?

There is no free route. Pick the one that fits how you work:

1. Gemini app: Subscribe to Google AI Pro or Ultra, open the Gemini app and choose Argon. Google's app announcement also added a chat-to-tasks toggle, so Gemini can keep working when you step away, plus connections to more than 40 apps such as Instacart, OpenTable, Adobe and Shopify.
2. AI Mode in Search: Pro and Ultra subscribers can use Argon inside Google's AI Mode.
3. Gemini API: Developers on the paid tier can call the model ID gemini-4-argon in Google AI Studio or the Gemini platform on Google Cloud.
4. Google Antigravity: Ultra subscribers get Argon in Google's agentic coding tool, which is now built into the Gemini macOS app.
5. Fairwind Program: Security teams can apply for a version without cyber guardrails.
6. Researchers: Google is offering 10,000 scientists six months of Antigravity's top usage tier free, subject to selection.

How much does Gemini 4 Argon cost?

For the API, Google lists an introductory price of $2 per million input tokens and $10 per million output tokens, with cached input 95% cheaper. After the introductory period ends on November 30, 2026, the price rises to $4 input and $20 output per million tokens.

At the introductory rate, Argon costs about the same as mid-tier models while benchmarking against flagships. Budget for the higher price, though. Argon can produce up to 1 million output tokens on a single task, and long agentic runs add up quickly.

How does Gemini 4 Argon compare to GPT-6 Astra and Claude Opus 5.5?

Argon leads on most of the benchmarks Google published, but not all. These scores come from Google's launch materials as compiled by andrew.ooo; all three labs choose and run their own evaluations, so treat them as directional.

Where Argon leads

• Real-world coding (DeepSWE v1.1): Argon 77.9%, GPT-6 Astra 74.1%, Claude Opus 5.5 74.2%.
• Long-context reasoning (GraphWalks): Argon 84.2%, Astra 71.8%, Opus 5.5 66.8%.
• Long video understanding (LVBench): Argon 91.7%, Astra 87.5%, Opus 5.5 83.7%.
• Finance agents (Vals Finance Agent v2): Argon 65.4%, Astra 53.5%, Opus 5.5 58.6%.
• Legal agents (Harvey Legal Agent Benchmark): Argon 19.6%, Astra 5.4%, Opus 5.5 3.8%.

Where rivals lead

• Frontier software engineering (FrontierSWE v2): GPT-6 Astra 65.5%, Argon 55.0%.
• Terminal tasks (Terminal-Bench 4.0): Claude Opus 5.5 66.4%, Argon 57.4%.
• Scientific terminal work (Terminal-Bench Science 0.1): GPT-6 Astra 68.1%, Argon 57.6%.
• Computer use: On the independent Vals leaderboard, Argon ranked only 7th of 8 on CUA-bench.

Independent rankings

On the independent Vals Index, which measures performance on economically important work, Argon placed first with 68.90%, narrowly ahead of Claude Sonnet 5.5 at 67.04% and Claude Opus 5.5 at 66.97%. The margins at the top are small.

What has Gemini 4 Argon done in real-world use?

Google shared several examples from its own teams and early partners:

• Chrome refactor: Argon helped the Chrome team separate core webpage handling from a legacy folder of about 6 million lines, finishing months of work in weeks.
• Faster code shipping: Inside Google, Argon in Antigravity increased agentic code submissions by more than 35% while cutting rollbacks by nearly a third.
• Faster open models: Argon delivered a 2.85x end-to-end inference speedup for Gemma 4 E2B on vLLM in a few days.
• Security: Wiz used Argon in its Scan for Good program to find and fix a flaw that had exposed a master key in healthcare software.

Is Gemini 4 Argon safe to use?

Google calls Argon its most thoroughly secured model so far. On the Gray Swan indirect prompt injection benchmark, attacks succeeded only 0.7% of the time across 15 adaptive attempts. Google also says shortcut attempts on coding tasks fell to 2%, and that a monitoring system reads the agent's reasoning to block unintended workarounds in real time.

What are the criticisms of Gemini 4 Argon?

Bloomberg reported on September 30 that some Google employees with direct access believe Argon scores well on benchmarks but struggles with real coding, especially front-end design. Google disputed that characterization. The independent results above point the same way: Argon is strongest on long, structured knowledge work and weaker on terminal and computer-use tasks. Test it on your own workload before switching.

What does Gemini 4 Argon mean for marketers and brands?

Argon now powers answers inside AI Mode in Search for paying subscribers. When the model behind an AI answer changes, which sources it trusts and cites can change with it, even when your Google rankings stay the same.

That makes three things worth doing this month:

1. Re-check your AI visibility. Run your most important category questions through Gemini with Argon and note whether your brand is named and which pages are cited. MarkGrid's Model Share tracks this across ChatGPT, Gemini, Perplexity, Claude and Copilot automatically.
2. Feed long-context models well. Argon handles very long documents, so deep, well-structured pages with clear headings and sourced data give it more to work with. Our GEO guide covers how to structure content for AI citation.
3. Pilot it on heavy analysis. Its strength on finance and legal benchmarks makes Argon a candidate for research-heavy marketing work, such as competitive teardowns and long report synthesis. Compare it against your current model on a real task before committing.

FAQ

Is Gemini 4 Argon free?

No. Argon is available to Google AI Pro and Ultra subscribers in the Gemini app and AI Mode, and to developers on the paid Gemini API tier. Google's announcements did not mention free-plan access.

What is the Gemini 4 Argon API model ID?

The model ID is gemini-4-argon, available on the paid tier of the Gemini API in Google AI Studio and on Google Cloud's Gemini platform.

How much does the Gemini 4 Argon API cost?

The introductory price is $2 per million input tokens and $10 per million output tokens, with cached input 95% off. From December 2026, after the introductory period ends on November 30, the price is $4 input and $20 output.

What is Gemini 4 Argon's context window?

Google says Argon supports a 1 million token unified input-output limit through the Gemini API, which allows very long generations in a single task.

Is Gemini 4 Argon better than GPT-6 Astra?

On Google's published benchmarks, Argon leads on most knowledge-work, long-context and video tests, while GPT-6 Astra leads on FrontierSWE v2 and Terminal-Bench Science. The best choice depends on your task.

What is the Fairwind Program?

Fairwind is Google DeepMind's program for vetted cybersecurity defenders. Members got Argon first and can access a version without cyber guardrails.

Last updated October 15, 2026.