Last updated: 14 October 2026, 5.46am BST.
The Gemini 4 Argon launch on 13 October 2026 brings Google’s new reasoning model to developers and paying Gemini app users. For Brighton businesses considering a trial, the immediate questions are access, running costs and reliability, not whether a launch-day benchmark proves it will improve their work.
In short
- Google announced developer and Gemini app availability on 13 October.
- The developer launch specifies the paid Gemini API and access routes through Google AI Studio; Ultra subscribers also have named access routes.
- Published introductory API rates run through 30 November 2026 and double on 1 December.
- DeepMind’s model card warns of hallucinations, slowness and timeouts.
Gemini 4 Argon launch: what is actually available
Google’s developer announcement positions Argon for lengthy software-engineering, knowledge-work and cybersecurity-defence tasks. Those descriptions are Google’s claims about the product, not Brighton Locals test results. The announcement points developers to paid API use and identifies Google AI Ultra access to AI Studio and Antigravity.
A separate Gemini app announcement names both Google AI Pro and Ultra subscribers. Meanwhile, the developer guide says Argon is not on the API free tier or the Google AI Pro plan. Do not treat the app subscription and developer access as interchangeable: check the entitlement for the exact product you intend to use.
The distinction is more useful than a blanket statement that Argon is either free or unavailable to Pro users. A team trying it in a chat window is making a different access decision from one integrating requests into its own software.
What the API prices mean
Google lists $2 per million input tokens and $10 per million output tokens through 30 November. From 1 December, the corresponding published rates are $4 and $20. Output charging includes thinking tokens. These are US-dollar API rates, not a quoted UK monthly subscription or an all-in project cost.
For a transparent illustration, one million input tokens plus 100,000 billed output tokens would total $3 at those introductory rates and $6 afterwards, before any caching discounts or other charges. That is simple arithmetic, not a prediction of what your application will consume. Retries and long answers make the actual workload important.
Specifications are not a verdict
Google’s model documentation lists text, image, video, audio and PDF inputs, with text output. It distinguishes a default synchronous output limit of 65,536 tokens from longer output using long decode. Image generation and audio generation are not supported by this model; computer use is labelled preview.
The DeepMind model card gives a March 2026 knowledge cutoff, while warning that some domains may be less current. Its benchmark table is a provider-published evaluation. It is not independent evidence that a particular Brighton agency, shop or research team will get the same outcome.
Reuters separately reported the launch on 13 October. Our technical specifications and access details above come from Google’s own announcements, documentation and model card, rather than treating that secondary news report as a technical authority.
Why it matters locally
A small team evaluating AI has to count the time spent checking its output as well as the API bill. A code migration that looks plausible but breaks a deployment is not a saving. This launch is distinct from our earlier UK–Canada AI security partnership report: it is a product release, not a government testing agreement.
Brighton Locals’ view
Argon merits a controlled trial, not an automatic switch. Clear published rates are helpful, but introductory pricing can hide the longer-term comparison. Our assessment would improve if independent testing, and a team’s own repeatable tests, showed fewer errors and less correction work at the December price.
What to do and watch next
Use the linked developer guide to confirm billing and supported features. Test a small, repeatable set of non-sensitive tasks, record errors and the time needed to correct them, and compare the bill at both published price levels. Keep purchases, deployments and changes to live customer data behind human approval while assessing the tool.
The next concrete cost milestone is 1 December. Recheck the official pricing and model documentation before a production commitment; features and availability can change after launch.
Bottom line
A new model is a useful option. Its real value is the work it can do reliably at the price you will actually pay.
Featured image and source
Archive: Googleplex headquarters building and Google sign, 27 July 2016. Photo: The Pancake of Heaven! / Wikimedia Commons, CC BY-SA 4.0. Not a screenshot of Gemini 4 Argon or its launch. Resized versions and display crops retain the same licence.
Leave a Reply