Gemini 4 Argon: Access, API Status and Confirmed Claims

Gemini 4 Argon has an official announcement, but that is not the same as a documented public API. Google announced the model on September 30, initially rolling it out through the Fairwind Program. On October 4, our checks found no Argon entry in the public Gemini API model list or pricing page, and no Argon listing in OpenRouter's public catalogue. That does not rule out private access. We have not called Argon or measured its performance. Here is what is confirmed, what an integration still needs, and how to interpret the announced prices without inventing a per-task bill.

If you arrived looking for a Gemini 4 Argon API key or a copy-ready request, there is an important missing step: a model name in a launch post is not an endpoint identifier. We will not turn a plausible ID into a working code example. The distinction also matters when comparing a vendor announcement with a model we have actually measured.

Gemini 4 Argon: confirmed announcement, unverified integration

The launch facts below come from Google's announcement, rechecked on October 4, 2026. Vendor-reported capabilities are not independent test results. The documentation checks are observations about the public pages we could read, not a claim to know every private deployment.

The useful distinction is between a confirmed product announcement and a confirmed integration path. Neither an attractive benchmark headline nor a plausible identifier fills in the missing API documentation.

Gemini 4 Argon public API status, checked October 4

We rechecked the Gemini API models page and pricing page on October 4. Neither had an Argon entry. OpenRouter's public catalogue returned 466 models and no identifier or display name containing argon or gemini-4.

These checks do not test private Google endpoints, Fairwind accounts or access for every enterprise customer. They establish a narrower point: the public integration sources we checked do not provide an Argon model entry. The spelling gemini-4-argon remains unverified here; we have not submitted a request to it and cannot report an error response.

What to verify before a Gemini 4 Argon integration

Prepare the evaluation now, but connect it only after the provider documents your access. A paid account, API key or Ultra subscription should not be treated as proof that the new model is enabled for that account.

1. Confirm the provider and account. Distinguish Google's developer API, a cloud deployment and a third-party gateway. Check which product the access announcement actually covers.
2. Copy the documented ID. Record the model entry, endpoint and SDK version from the provider's own documentation. Keep any account-specific access conditions with the record.
3. Read the applicable price. Note introductory terms, input and output units, cache charges and other fees before approving a test budget.
4. Make an authorized smoke test. After access is confirmed, use a small non-sensitive request. Save the returned model identifier, usage, latency and charge; a successful request still does not demonstrate the advertised capabilities.
5. Run the same acceptance set. Use tasks and pass criteria fixed before testing. Include errors, retries and a documented fallback. Compare the result with your current model rather than with launch graphics.

Until those steps are possible, an existing model with documented access is the operational choice. Our Gemini API key guide covers credentials; a key is not a substitute for model access.

What $2 / $10 buys today, measured

We have not run Argon. The table compares its announced input/output prices with other models measured on our nine Python tasks. Their prices were captured on 2026-10-02; the table is not a live catalogue. GPT-6 Sol Pro is excluded because its recorded input token count differs substantially. The comparison covers the uncached input/output rates, not identical cache terms, access or every possible fee.

All four costs are derived from measured tokens times the list price captured 2026-10-02. On an identical rate card these three $2 / $10 models land between $1.61 and $2.03 per thousand tasks, because they write different amounts. The rate card sets the price of a token; the model decides how many you buy.

The table gives you a budget reference, not an Argon score. At the announced post-introductory input/output prices, Argon would share those two rates with the dated Opus 5.5 record. It would not inherit Opus's token use, latency or task cost. Likewise, the three models at $2 / $10 passed these nine functions, not every task an application might need.

And to be plain about what our suite can show: nine self-contained Python functions cannot separate a frontier model from a competent small one. The cheapest and fastest 9 out of 9 among the 75 models that scored it is Solar Mini 4, at $0.03 per thousand and 2 seconds (priced 2026-10-02). When Argon is callable, this suite will tell you its floor price and its verbosity. It will not tell you whether DeepSWE 77.9% is real.

Google's last price move, for calibration

A prior Flash run is useful for understanding why a rate card does not determine task cost. It is not a reliable prediction of Argon's behavior. Our two dated Gemini Flash runs show separate effects from token use and price:

The $1.40 figure belongs to the August 22 price capture, not to the current Gemini price list. Our Gemini 3.8 Flash article keeps the dated price and token-use comparison together. The two models and capture dates differ, so this is not a controlled before-and-after experiment on one unchanged endpoint.

Gemini 3.8 Flash wrote 9,168 output tokens across the suite; GPT-6.1 Sol wrote 1,329. That is about 6.9 times as many, but it does not tell us how much Argon will write. An output limit is a ceiling, not a token-use prediction. Estimate Argon's task cost only after observing its own requests and the rates actually charged. For family-level billing distinctions, see our Gemini API pricing guide.

How our numbers were produced

Nine Python tasks, each a function signature and a spec. Generated code runs against hidden asserts in an isolated python3 -I subprocess with a 12-second timeout. Temperature 0, max_tokens 4000, one scored attempt per task. An API-layer failure is recorded separately from a wrong answer. Cost is derived: measured input and output tokens times the list price on the capture date shown beside each figure. It is not a billing statement. Runs go through OpenRouter. Updating the public API status on October 4 did not rerun the benchmark or test Argon. Full method on the methodology page; for why every price above carries a date, see our note on price volatility.

What we have not verified

• When Argon opens up. Google says paid API customers and AI Ultra subscribers come first. It gives no date.
• The model ID. Not verified in the public integration documentation we checked. Private access is outside this review.
• The input context window. The announcement's 1M output limit should not be substituted for an input context specification.
• How long $2 / $10 lasts. The announcement calls it introductory and states the later $4 / $20 prices. We have not verified a start or expiry date for a public account.
• Its token habits, and so its real per-task cost. That needs a run, and we have not done one.
• Private access. We did not obtain Fairwind access, test a private endpoint or inspect another customer's account.

Gemini 4 Argon FAQ

Has Google actually announced Gemini 4 Argon?

Yes. Google's September 30 announcement is a primary source. That confirms the announcement, not that every developer can already call the model.

Can I call gemini-4-argon with my existing key?

We have not verified that ID or public account access. On October 4 the public Gemini API model and pricing pages had no Argon entry. Wait for your provider's documented identifier and account eligibility rather than relying on this display-name spelling.

What does Gemini 4 Argon cost per task?

We cannot give a measured figure. Google announced introductory input/output rates and later rates, but a task estimate needs Argon's own token usage and applicable fees. The other models in this article do not stand in for an Argon run.

Prepare the evaluation, not a guessed endpoint

Gemini 4 Argon is a model to evaluate when documented access is available to your account. For now, keep the provider, ID, price terms and acceptance tests ready. The missing evidence is an authorized Argon run, not another launch summary.

Evidence: dated measurements and catalogue fields; method, latency statistics and limitations. Measurements describe these runs, not all providers or future versions.

Primary sources, checked October 4, 2026: Google announcement and pricing footnote; Gemini API model documentation; Gemini API pricing; OpenRouter public catalogue. The benchmark measurements retain their original capture dates.