Gemini 4 Argon pricing: intro rates run to Nov. 30 as early rankings shift

As of 9:10 a.m. ET, Tuesday, Oct. 20, 2026

Google's Gemini 4 Argon costs $2 per million input tokens and $10 per million output tokens through Nov. 30, 2026, and those rates double to $4 and $20 starting Dec. 1, according to Google's Gemini API pricing page, last updated Oct. 19. Google announced the model's general availability on Oct. 13.

A week later, the independent ranking Google pointed to at launch has changed. Google's Oct. 13 announcement said Argon ranked No. 1 on the Vals Index. Vals AI's own leaderboard, updated Oct. 19, now lists it third.

What it costs now, and on Dec. 1

On the Gemini Developer API's standard paid tier, Google lists input at $2.00 per million tokens through Nov. 30 and $4.00 from Dec. 1. Output, which Google says includes "thinking" tokens, is $10.00 through Nov. 30 and $20.00 after.

Cached input is $0.10 per million tokens through Nov. 30 and $0.20 after, Google says. Argon supports only implicit caching, which Google says discounts cached input by 95%.

The same pricing page lists Batch and Flex rates of $1.00 in and $5.00 out through Nov. 30, rising to $2.00 and $10.00. Priority processing is listed at $3.60 and $18.00 through Nov. 30 and $7.20 and $36.00 after.

On Google Cloud's Agent Platform, prices on the global endpoint match the API's, and non-global endpoints are listed 10% higher (for example, $2.20 and $11 through Nov. 30). Google Cloud says the promotional price is provided "through 50% credits back on net spend" and does not apply to a separate version it calls Gemini 4 Argon Cyber Permissive.

The 1-million-token limit

Google's developer documentation says input, thinking and output share "one window of 1,048,576 tokens." It says a single response can now run past 64K tokens, up to 1 million tokens minus the input, a feature Google calls long decode.

The default output limit is 65,536 tokens. The documentation says the easiest way to get longer answers is to turn on background execution and raise the output limit; developers can also run the continuation themselves.

Google's launch post describes the 1M limit as available "via the Gemini API." Google Cloud's pricing page says Gemini 4 can generate completions up to 1 million tokens in length. The Google pages GROKsiNEWS.com reviewed did not state a limit for the Gemini app or AI Mode.

Google Cloud's pricing page also lists a billing change. When a completion runs past 32K tokens, users "will be charged for retrievable token usage even if the request ultimately fails due to client or server timeouts," the page says. It says this takes effect Nov. 13, 2026.

Who can use it

Argon is not available on the Gemini API's free tier, Google says. According to Google's launch post, developers can build with it on the paid API tier in AI Studio and Android Studio, and Google AI Ultra subscribers can try it in Google Antigravity and AI Studio.

Google says Google AI Ultra and Pro subscribers can use Argon in the Gemini app and in AI Mode in Search. Enterprises can use it in Gemini for business, the API and Antigravity. Google Cloud said Oct. 13 that Argon is available in Gemini for business on the web, desktop and mobile, and through its Gemini platform under the model ID gemini-4-argon.

Google also says it will offer 10,000 scientists its highest Antigravity usage tier free for six months, subject to eligibility, capacity and selection. Cyber defenders can apply to its Fairwind Program for access to Argon without cyber guardrails.

How it stacks up, by who ran the test

Vals AI's Vals Index, updated Oct. 19 to version 2.2, ranks Anthropic's Claude Sonnet 5.5 first at 70.85%, Claude Opus 5.5 second at 70.18% and Gemini 4 Argon third at 69.24%. Vals AI says some component tasks for the two Claude models were served by a fallback model after a provider refusal. Counting those as failures would lower their scores to 69.03% and 64.16%, and because Argon had no fallbacks, "it would rank first," Vals AI says.

Vals AI's Sept. 30 write-up, made before the Oct. 19 version update changed several of the index's components, had placed Argon first of 41 models at 68.90%.

Artificial Analysis said in a Sept. 30 write-up that Argon, run at high reasoning, scored 53 on its Intelligence Index, matching OpenAI's GPT-6 Astra. Its model page, checked Tuesday, lists Argon at 53, eighth of 217 models on that index.

In the same write-up, Artificial Analysis reported a 15% hallucination rate on its AA-Omniscience test, the lowest it found among models scoring 45 or more on its index, and a 57% Terminal Bench 4 score that trailed three rivals.

On LMArena's Text Arena, where users vote between anonymous answers, Argon ranks first with a preliminary score of 1524, in an update dated Oct. 15. LMArena lists its possible rank range as first to third.

What Google itself claims

Google Cloud's launch post says Argon "leads on the Vals Index" and on Harvey's Legal Agent Benchmark. Google's launch post reports an 87% score on the Embodied Reasoning benchmark ERQA and says Argon "outperforms all other frontier models" there. Those are Google's own figures.

None of the Google pages GROKsiNEWS.com reviewed announced an extension of the Nov. 30 introductory pricing.