Gemini 4 Argon undercuts rivals on price, but only until 1 December

Google is pricing Gemini 4 Argon to win developers early, and the discount has an expiry date. Introductory API rates of $2 per million input tokens and $10 per million output tokens double from 1 December. Google opened the model to paying users and developers worldwide on 14 October, just under two weeks after introducing it. Until now, only selected test partners had access, according to heise online. Other outlets date the rollout to 13 October, and the introduction date of about two weeks earlier was not independently checked.

Access runs across Google's whole stack. Google AI Pro and Ultra subscribers get Argon in the Gemini app and in Google Search's AI mode. Developers pay for it through the Gemini API in AI Studio, and companies get it through Google Cloud. Ultra subscribers also get it in Antigravity.

Google is also courting researchers. Up to 10,000 selected scientists can use Antigravity's highest usage tier free for six months, but they must apply. Cached input tokens cost 95 percent less than standard input tokens, which favours workloads that reuse long prompts. The context window is one million tokens in total, covering input, internal calculations and output.

On benchmarks, Argon is competitive rather than dominant. At its highest reasoning tier it scores 53 on the Artificial Analysis Intelligence Index. Claude Opus 5.5 (Max) leads with 58, while Claude Sonnet 5.5 (Max) and Meta's Muse Wonder Full (Max) each score 56.

It does better on agent-style work. In AutomationBench-AA, Argon met 78 percent of task goals without rule violations. Gemini 3.8 Flash (High) managed 60 percent and Claude Opus 5.5 (Max) 70 percent. In AA-Omniscience, Argon answered 50 percent correctly and was clearly wrong on only 15 percent of the remainder, often giving no answer or a partial one instead. Artificial Analysis calls its hallucination rate comparatively low.

Cost per task is where the pricing shows. The average cost per Intelligence Index task is $1.99 at introductory prices, and about $3.98 without the discount. Artificial Analysis put it this way: "The comparatively low costs are primarily due to the temporarily favorable token prices".

Google is trading margin for adoption. Argon trails the top Anthropic and Meta models on general intelligence, so a low entry price is the lever for getting developers to build on it before they compare it with rivals. The 95 percent cache discount and the free scientist tier push the same way: lock in heavy users early.

The exposure is in December. Doubling the rates doubles the per-task cost the benchmarks currently show, from $1.99 to about $3.98. Teams that pick Argon on price will face a repricing just as they move to production. Operators should budget at the higher figure and test whether Argon's lead on automation tasks, 78 percent against Opus 5.5's 70 percent, justifies it. Watch whether workloads stay after the discount ends.