Gemini 4 Argon: API pricing, Pro and Ultra access, and test results

👁 115 reads

Google has opened Gemini 4 Argon to all paying users. Since Tuesday, October 13, 2026, its flagship model has been available through the paid Gemini API, in the Gemini app for Google AI Pro and Ultra subscribers, in AI Mode in Search, and in Antigravity for Ultra subscribers. The launch API price is $2 per million input tokens and $10 per million output tokens through November 30. A few hours after the announcement, I ran my tests: 99.5% across 23 practical tasks for $0.22, 81.8 points in Polish-language writing, and a perfect 220 out of 220 on Poland’s school-leaving exams. I collect the details in my Gemini 4 Argon guide.

What Google announced

At 18:04 Polish time, Google CEO Sundar Pichai wrote on X: “Gemini 4 Argon is here. It’s available globally today for developers and enterprises in the paid tier of the Gemini API and the Gemini platform, for Ultra subscribers in Google Antigravity, and Ultra and Pro subscribers in the Gemini app and Search.” A short time later, Google published two blog posts, one about the model and one about the Gemini app. Here are the key details from both:

• Who can use it: developers and businesses through the paid API (Google AI Studio and Gemini platform); Google AI Pro and Ultra subscribers in the Gemini app and AI Mode in Search; and Ultra subscribers in Antigravity, both the app and CLI.
• Who cannot: users on the free plan and Google AI Plus. Those plans remain on the 3.5 Flash-Lite and 3.6 Flash models.
• Researchers may be able to get the highest Antigravity tier free for six months; applications are accepted on a rolling basis.
• Fairwind program: the program, which has given cybersecurity specialists access to Argon since September 30, is expanding to “hundreds” more partners. They receive a version without safeguards against use in attacks.
• In the Gemini app, the model picker will have four options: 3.5 Flash-Lite, 3.8 Flash, 3.1 Pro and 4 Argon. Google says Argon gives more direct answers, hedges less and is less repetitive.

Gemini 4 Argon API pricing: $2/$10 through November 30

After the promotion, Argon will cost the same as Claude Opus 5.5 and Muse Wonder ($4/$20); through the end of November, it costs the same as GPT-6.1 Sol. GPT-6 Astra and Claude Fable 5.1 cost $10/$50. One important correction to what I wrote on September 30: Google now clarifies that the one-million-token limit in the Gemini API is shared between input and output, not one million output tokens on top of the context window.

My first tests: inexpensive, fast and strong in Polish

Argon appeared on OpenRouter alongside the announcement, so I immediately ran three test sets, all at low reasoning effort:

• 23 practical tasks (an agent using tools, code checked with tests, Polish-language documents, reasoning and a 158,000-character document): 99.5% for $0.22 in 135 seconds. The same test set gave Muse Wonder 99.7% for $0.64 on October 12, Claude Opus 5.5 99.4% for $0.75, and GPT-6 Astra 100% for $1.12. Argon was the cheapest and fastest flagship in this test.
• Polish-language lab (13 tasks, scored blind by Opus 5.5 and GPT-6 Sol): 81.8 points. This is the best result for a Google model in my ranking (Gemini 3.8 Flash scored 79.4; Gemini 3.1 Pro, 77.8), but Argon placed only eighth overall. Opus 5.5 scored 90.1 and Muse Wonder 87.0. We tested it in Polish; this is not an English-language or multilingual benchmark.
• May 2026 school-leaving exams: 50/50 in basic mathematics, 50/50 in advanced mathematics, 60/60 in Polish and 60/60 in English. A perfect score, matching Opus 5.5 and GPT-6.1 Sol. The Polish-language portions were tested in Polish.

The only point deducted came from a mistake by my automated grader. In a test involving an appeal of a Polish social insurance agency decision, the script looked for the phrase “district court,” while Argon correctly wrote “To: the District Court in Kraków.” This English wording is a translation of the original Polish test material; the actual test was conducted in Polish. After manual review, Argon would have received full marks. The more substantive issue with the letter was that it added symptoms I had not included in the prompt. In a test of fabricated sources, it returned 11 of 16 genuine publications with correct details and made up two sources, but did not invent any DOI numbers. It counted the same Polish document as 51,482 tokens, compared with 87,675 for Claude.

What independent rankings say

• Artificial Analysis: 52.6 points on its Intelligence Index, effectively tied with GPT-6 Astra (52.7) and behind Claude Opus 5.5 (57.6). Argon had the lowest hallucination rate among the leading models: 15%, compared with 51% for Astra. Running the full test set costs $1.99 at the launch price, compared with $3.26 for Astra.
• Arena (formerly LMArena): first place on the text leaderboard with a provisional score of 1,525 points, and eighth place in Code Arena: WebDev.

Google also shared new figures, including 87% on ERQA, a benchmark for robotic reasoning, and a 0.7% successful attack rate on Gray Swan IPI. These are figures reported by Google. The Antigravity team added that, in Google’s code, Argon produced 35% more agent-written code and 30% fewer rollbacks after deployment - also figures reported by Google.

How to access Gemini 4 Argon

The quickest route is the API: use a paid account in Google AI Studio or OpenRouter, and Argon is available without waiting for the app rollout. Pro and Ultra subscribers should check the model picker in the Gemini app, since Google is enabling the model in stages. Antigravity users need Ultra. My tests are covered in the 23-task test and the Polish-language Gemini 4 Argon test; API costs are in the Gemini 4 Argon pricing article, and this guide covers what to do if the model is missing from the app. For an overview, see the Gemini 4 Argon guide. I covered the first reactions on September 30 in the launch article. Availability confirmed in Poland on October 13 does not establish availability in every country; check Google’s availability and rollout for your location.

Frequently asked questions

What is Gemini 4 Argon?

Gemini 4 Argon is Google’s flagship language model, available to paying users since October 13, 2026. You can use it through the paid Gemini API (model ID gemini-4-argon), the Gemini app on Pro and Ultra plans, and OpenRouter. It ranked first in Arena’s text category. In practical tests, it achieved 99.5% accuracy and earned a perfect score on Poland’s school-leaving exams, while also being the cheapest and fastest flagship in this test set.

How much does the Gemini Argon API cost?

Through November 30, 2026, Gemini 4 Argon costs $2 per million input tokens and $10 per million output tokens. From December 1, the prices double to $4 and $20. Cached input costs $0.10 through November, then $0.20. The one-million-token limit covers input and output combined. API access requires a paid account in Google AI Studio or OpenRouter.

What is the difference between Gemini Argon on Pro and Ultra?

Both plans provide access to Gemini 4 Argon in the Gemini app and AI Mode in Search. Antigravity, Google’s agentic coding tool, is available only on Ultra. The app rollout is phased, so Argon may not appear in the model picker immediately; access may arrive later. Check Google’s current plan and availability details for your location.

// read moreGemini 4 Argon in the App: Rollout Status and ChecklistNewsGoogle Gemini 4 Argon: New Model for Complex Professional TasksNewsIs Gemini 4 Argon Free? Access Options and Costs ExplainedGuides

Kinetyra