Gemini 4 Argon's $2 Introductory Price Doubles on November 30
A.I. / news
Gemini 4 Argon's $2 Introductory Price Doubles on November 30
Google's launch post and a newsletter written days earlier disagree on who can use the model, and the independent scores sit alongside a hallucination figure that only one of them reports.
Google's Gemini 4 Argon costs $2 per million input tokens and $10 per million output tokens until November 30, 2026, after which the public price becomes $4 and $20, according to Google's launch post.
The post says the model is available worldwide from launch to developers through Google AI Studio and the Gemini API, and to Google AI Ultra and Pro subscribers in the Gemini app and AI Mode in Search. The Gemini API documentation lists the model ID as gemini-4-argon and calls it "now available".
That is not what The Batch, the newsletter from DeepLearning.AI, reported. Its piece, dated October 9, said Argon was "currently only available to organizations in Google's Fairwind cybersecurity program" and did not give a date for wider access. The Batch also said the end date of the introductory price was undisclosed. Google's post gives November 30. The launch post carries no date in the copy The Terminal fetched, so the order of events is inferred from the newsletter's date. Google did not explain the gap.
What the model costs
The API documentation page does not list prices. The launch post and The Batch agree on the introductory figures.
| Token type | Until Nov. 30, 2026 | After Nov. 30, 2026 |
|---|---|---|
| Input, per million | $2 | $4 |
| Output, per million | $10 | $20 |
| Cached input, per million | $0.10 | Not stated |
Google said cached input tokens are discounted by 95%. The $0.10 cached figure comes from The Batch. Both prices apply to a 1,048,576-token context window shared between input and output, per the documentation. Output defaults to 65,536 tokens and can reach 1,048,576 with the API's long-decode option.
The Batch put the standard-price cost of one Artificial Analysis task at $3.98, against $1.99 at the discounted rate. It said that places Argon level with GPT-6 Astra and Claude Fable 5.1, both at maximum reasoning, on that index at a lower cost per task. The price rise on November 30 removes the discount from that comparison.
What independent scores say
Google's own post says Argon ranks first on the Vals Index. The Batch reported the figure behind it: Vals Index v2.1, 68.9% accuracy, first of 43 models at high reasoning. The index covers eight finance, coding, legal and tax benchmarks.
On the Artificial Analysis Intelligence Index v4.3.2 the model scored 53 at high reasoning, tied for third, according to The Batch. The same piece said it still trails Anthropic's newest models on agentic coding. On Arena.ai's Text Arena it held first place at 1,525 Elo, ahead of Claude Opus 4.6 at 1,505 and Claude Opus 5.5 at 1,504.
- Gemini 4 Argon (high reasoning)15 %
- GPT-6 Astra (max reasoning)51 %
Source: The Batch (DeepLearning.AI), Oct. 9, 2026, citing Artificial Analysis
The Batch said Argon's 15% is the lowest among models scoring 45 or higher on the Intelligence Index. Google's post does not report this figure. It reports its own: 87% on ERQA, an embodied-reasoning test, 57% on Gamow Labs' RareBench for rare-disease diagnosis and a 0.7% attack success rate on Gray Swan's indirect prompt injection benchmark across 15 adaptive attempts. Those are Google-supplied and The Terminal has not seen independent runs.
What Google claims it did with it
The post says Argon helped the Chrome team separate core web-page functions from a legacy directory of roughly 6 million lines inside Chromium's 40 million, across hundreds of files, in weeks rather than months. It also says Google scientists used it to produce five serine esterase enzyme designs in hours, and that lab tests confirmed catalytic activity. Both are Google's accounts. The post names Wiz, Isomorphic Labs, Gamow Labs and Apptronik as partners and carries the name of Koray Kavukcuoglu, SVP of Google DeepMind and Chief AI Architect at Google.
On safety, Google said approved cyber defenders can apply to the Fairwind program for access without cyber safeguards, and that the program is expanding to hundreds more partners. The Batch said the broader release refuses harmful cyber and chemical, biological, radiological and nuclear requests.
For another OpenAI pricing reference point, see OpenAI's Decisions API pricing. The same caution about self-reported scores applies to Laminar's Flow-1 launch, where the vendor's own table showed a competitor ahead.
Google said researchers can apply for six months of free access to the top usage tier of Antigravity, its coding product. The introductory price ends at the close of November 30.
Sources
More in A.I.
- 01Lean Creator Leo de Moura Says AI Makes Proofs Cheap. Here Is What the Checkers ShowOn a podcast released October 29, the AWS scientist who built Lean claimed AI-written proofs now run to 50 million lines. The claims check out only in part, and he names the two gaps himself.
- 02Claude's Watermark Quality Claim Rests on Single Responses, Blogger SaysAnthropic says watermarking does not affect output quality. A post citing Google's own SynthID-Text paper says the tests never measured what happens when you ask the same question twice.
- 03Nanbeige4.3-33B Ships Under Apache-2.0 With a 524,288-Token Window and Self-Reported WinsNanbeige Team's 33B model is built as a teacher for smaller agents. Its own table has Qwen3.8-27B ahead on SWE-Bench Pro and GDPval.
- 04GPT-5.6 Sol Launched a Factorio Rocket on Peaceful Mode, With Hauling Done by HandBlogger Ryan Madden's single run took 13.5 in-game hours. The agent carried supplies between areas because it never linked some production lines.