Model News

Gemini 4 Argon API: Access, Limits and Pricing

Gemini 4 Argon is now listed in Google's public Gemini Developer API directory as gemini-4-argon. That changes our October 12 documentation finding. The remaining checks are your paid account's eligibility, the request mode and its limits, and the applicable price. No Argon inference or account-level access test was run for this revision.

Gemini 4 Argon public API, Standard token rates through November 30, 2026, and documentation-only scope

Gemini 4 Argon has a documented public API ID

On October 15 we checked the model directory, Argon model page and price table. They now identify gemini-4-argon. The earlier absence on two checked pages was a dated observation, not proof that access could never be public.

The paid-tier requirement still matters: the price table excludes the Free tier and Google AI Pro. An API listing does not certify an individual account, region, gateway or consumer subscription. Check the route you intend to use; do not carry an invited-program assumption into a different product.

Normal output and long decode are different request modes

The model page specifies a unified 1,048,576-token context and a 65,536-token default synchronous output limit. Output up to 1,048,576 requires the separately documented long-decode path. Do not read the launch headline as one million output tokens available on every normal request. Source and long-decode link

A model ID, context ceiling and completed answer are three different checks. A server can accept a request and still return a truncated, blocked or incomplete result. Keep the finish reason and usage with any future test. A default token limit is not a target output size.

Dated Gemini 4 Argon Standard pricing

The official Standard paid-tier USD rates are shown below. Batch, Flex, Priority and tool use have separate terms; this is not an all-route quote. Google price table

Per million tokensThrough Nov 30, 2026From Dec 1, 2026
Input$2$4
Output, including thinking$10$20
Implicit cache input$0.10$0.20

The table says implicit caching only, not explicit cache creation. Keep that distinction when reusing a previous Gemini integration. Tool charges and failed attempts can add cost; no per-task Argon cost or invoice was measured here.

Prepare one account-level check, not a guessed integration

Record the following before approving a small non-sensitive request:

Product: Gemini Developer API, not an assumed Vertex deployment
Model: gemini-4-argon
Account and region: confirm actual eligibility
Request mode: normal synchronous or separately documented long decode
Rate date and billing tier: retain the applicable table
Budget: declare maximum spend, output and elapsed time
Result: record request ID, returned model, finish reason and full usage
Disposition: accepted / failed / inconclusive; no silent fallback

First establish the documented route and account permission. Then verify one completed response and its returned identity before involving private files or a tool loop. For credential setup, use the Gemini API key guide. That guide addresses a key, not permission for every model.

A paid model request can incur charges. No key was used and no generation request was sent while preparing this article. There is consequently no first-party speed, coding-quality or tool-reliability result to publish. Keep vendor capability statements separate from an actual workload evaluation.

What changed from the earlier access note

The October 12 check is retained as history: no Argon entry was found in the two Developer API pages read then. The October 15 public documents supersede that access-status conclusion and provide both an identifier and dated rates. Neither snapshot says what a particular account successfully called.

The September 30 announcement remains launch context. Use current integration documentation for request settings. The testing methodology explains why another model's archived results cannot stand in for an Argon run.

Conclusion: establish the documented route, then test your workload

Gemini 4 Argon now has a public API identifier and a dated paid-tier price schedule. Verify your account and request mode before testing, and judge completed work rather than a launch headline. This revision establishes documentation changes, not successful inference.

FAQ

Is gemini-4-argon still a guessed API ID?

No. It is now listed in the official Gemini API directory and model page. Actual account access was not tested here.

Can a normal request generate one million tokens?

The model page lists 65,536 as the default synchronous output limit. Larger output uses the separately documented long-decode path.

Was an Argon performance test run for this update?

No. This revision checks public documentation and pricing; it reports no model run, latency, capability score or actual bill.

Written by

Prepared with AI assistance. This revision checks documentation, consolidates dated records and tests offline examples. It does not claim new model, integration or hardware runs.