Gemini 4 Argon Arrives for Google AI Pro: Pricing, API Rates, and What Stays Ultra-Only

Gemini 4 Argon is rolling out to Google AI Pro in the Gemini app, while Antigravity and AI Studio stay Ultra-only. Availability by tier, API pricing of $2/$10 per million tokens before it doubles, and Google's reported engineering results.

PUBLISHED // UPDATED
DURATION // 07 MIN READ
Laptop showing a paid-tier model picker next to a printed $2/$10 token rate card on a desk, cool daylight, no logos
FIG.01 // VISUAL DOCUMENTATION Laptop showing a paid-tier model picker next to a printed $2/$10 token rate card on a desk, cool daylight, no logos

Google’s Gemini 4 Argon is now rolling out to Google AI Pro subscribers in the Gemini app, a month after it first appeared for AI Ultra subscribers only. The release is staggered across subscription tiers and products, and its API pricing comes with an introductory discount that changes later in the year. Knowing which tier unlocks which surface, and how the API price will change, matters more for planning than the benchmark claims in the launch coverage.

This article summarizes availability by tier, API pricing, the engineering examples Google has shared, and the security measures it describes, with practical guidance for users and developers.

Who gets Argon, and where

9to5Google reported on October 13 that Gemini 4 Argon is rolling out in the Gemini app to Google AI Pro and AI Ultra subscribers, after an initial Ultra-only release the previous month. Google describes it as offering “frontier reasoning and improved performance across complex workflows.”

The Verge adds that Argon is available in the Gemini app and in AI Mode in Search for Pro and Ultra subscribers, and in Antigravity, Google’s agentic development platform, for Ultra only. It was not clear whether lower tiers or free users would get access. The Next Web reported that Google AI Pro starts at $19.99 per month and that Google had moved free users to its smallest model a few days earlier.

SurfaceAI Pro ($19.99+/month)AI Ultra ($99.99+/month)
Gemini appYesYes
AI Mode in SearchYes, per The VergeYes
Antigravity (agentic development)NoYes
AI Studio and Android Studio CanaryNo, per 9to5GoogleYes

In practice, Pro unlocks Argon for chat and documents, while coding and agent workflows remain an Ultra feature. Availability may vary by account, so users should confirm the model is selectable rather than assume it.

API pricing and the discount date

9to5Google lists introductory API pricing of $2 per million input tokens and $10 per million output tokens until November 30, 2026, rising to $4 and $20 afterwards. The Verge estimated the introductory price at around a fifth of the cost of OpenAI’s GPT-6 Astra and Anthropic’s Fable 5, and said Argon should remain cheaper per token even after the increase. It also cautioned that per-token prices do not equal per-task costs, because models use different numbers of tokens for the same task.

The Next Web reported the same prices but described the discount as running “at least a month with no stated end date,” which conflicts with 9to5Google’s November 30 date. Teams planning significant usage should confirm the current rate in their Google Cloud billing account and model both price levels.

Subscription prices and API prices are separate. A Pro subscription covers use in the app; it does not include API usage. An Ultra subscription unlocks Antigravity but is not an API token allowance.

The engineering examples Google shared

Google highlighted several internal uses, all reported by the company:

  • YouTube comments. According to 9to5Google, Argon running in Antigravity analyzed production profiles, found that redundant emoji rendering accounted for nearly 45% of server overhead, and produced 12 targeted optimizations that cut server comment-load latency by 30 ms and client end-to-end latency by 20 ms. Ars Technica summarized this as a 45% drop in server load; The Verge said Argon halved the compute for the comments section.
  • Inference optimization. 9to5Google quotes an example in which Argon optimized inference for the Gemma 4 E2B model on an A100 with vLLM, achieving 2.85× end-to-end speed (about 37,900 to over 108,000 tokens per second) with identical output, in a few days.
  • Security and simulation. The Verge reports Argon was used internally to find and patch security vulnerabilities and to debug the drone physics simulator at Alphabet’s Wing; Ars says some of those simulator bugs had existed since 2018.

These examples show what the model can do with access to detailed production data and expert engineers who review its changes. They are not a guarantee of similar results elsewhere. Teams hoping for comparable gains need performance profiles, permission to change the relevant code, and reviewers who can judge the proposed patches.

Benchmarks and early access

Google says Argon ranks first on the Vals Index for finance, coding, legal, and tax work, and The Verge reports Google’s claim that it outperforms GPT-6 Astra, Claude Fable 5.1, and Claude Opus 5.5 across several domains. These are company-reported results. Organizations comparing models should test them on their own tasks.

The Next Web reported that Argon’s first external users were cyber defenders who received the model without its usual guardrails through a Google program called Fairwind, which is expanding toward hundreds of partners. TNW reported that security firm Wiz used the model to scan critical infrastructure and found a bypass exposing a master key in healthcare software; readers should look for Wiz’s own disclosure for details. Tulsee Doshi, Gemini product lead at Google DeepMind, described Argon to Business Insider as the first of a family of models, according to TNW.

Safety measures and the delayed release

Ars Technica reports that Google revealed Argon weeks earlier but delayed wider release over concerns about “nefarious use.” According to Ars, the model’s capabilities did not change from the September reveal; it is trained to refuse malicious instructions, and an automated misalignment monitor can halt a workflow. Ars also notes that thousands of Google employees have used the model internally.

These are Google’s descriptions of its safeguards. Security teams should ask what the monitor logs, who is notified when a workflow is halted, and whether any contracts include access without guardrails. TNW notes that offensive cyber capabilities are treated as a systemic-risk category in Europe, which may matter for organizations subject to EU rules.

Practical guidance

For AI Pro subscribers: check whether Argon is selectable in the Gemini app. Do not assume AI Mode in Search has it just because the app does.

For developers using the API: run ten representative tasks, log input and output tokens, and calculate cost at both $2/$10 and $4/$20. If the use case only works at the introductory price, plan for the change.

For teams considering Antigravity: budget for Ultra, confirm Argon is enabled for your organization, and start with one repository that has good tests and performance profiling, as Google’s YouTube example did.

For Android developers: Argon in Android Studio requires the Canary channel and Ultra, according to 9to5Google. Demos on Canary may not reflect what stable releases offer.

For free-tier users: Argon is not currently available, and TNW reports free users have been moved to a smaller model.

See our earlier analysis of Gemini 3.8 Flash and our coverage of self-reported coding benchmarks, which explains why vendor-reported results should be checked independently.

Bottom line

Gemini 4 Argon is now available to AI Pro subscribers in the Gemini app, while its coding and agent features in Antigravity, AI Studio, and Android Studio remain limited to Ultra. API pricing starts at $2/$10 per million tokens and is scheduled to double, with sources differing on when. Google’s engineering examples are impressive but company-reported. The sensible approach is to confirm access for your tier, test cost at both price levels, and evaluate the model on your own work before committing.

Sources