Gemini 4 Argon Pricing: What the $2 Intro Rate Gets You

Gemini 4 Argon is now broadly available, and Google has attached an introductory price that makes its flagship model considerably cheaper to try. In a post dated October 13, 2026, Google DeepMind's Koray Kavukcuoglu said Argon is open to developers, enterprises and paying subscribers, two weeks after it debuted with a restricted group of security teams.

• Intro pricing: $2 per million input tokens and $10 per million output tokens through November 30, 2026.
• Standard pricing: $4 input and $20 output per million tokens after the introductory period.
• Caching: cached input tokens are 95 percent off.
• Scale: a 1M unified input-output token limit through the Gemini API.

How Much Does Gemini 4 Argon Cost?

Google's announcement sets the introductory rate at $2 per million input tokens and $10 per million output tokens. When the introductory period ends on November 30, 2026, the price becomes $4 and $20. Cached input is discounted by 95 percent, which VentureBeat calculates as $0.10 per million tokens at the introductory rate.

That discount matters for anyone running agents. Long-running agent sessions resend the same instructions and documents many times, and caching is where most of that repeated cost can be removed. Teams evaluating Gemini 4 Argon should measure their cache hit rate before estimating a monthly bill.

VentureBeat's price comparison places the introductory rate level with Claude Sonnet 5.5 and GPT-6.1 Sol, and the standard rate level with Claude Opus 5.5. If you are weighing those tiers, our guide to Claude Sonnet 5.5 versus Opus 5.5 walks through the same trade-off.

Who Can Use Gemini 4 Argon Now?

According to Google, access now covers four groups:

• Developers: the paid tier of the Gemini API, in AI Studio and Android Studio. The API changelog lists the model ID as gemini-4-argon.
• Enterprises: Gemini for business, the API and Google Antigravity.
• Subscribers: Google AI Pro and Ultra members get Argon in the Gemini app and in AI Mode in Search, and Ultra members also get it in Antigravity.
• Scientists: 10,000 researchers will receive Antigravity's highest usage tier free for six months, subject to eligibility, capacity and selection.

Google is also widening the Fairwind Program, the track for security teams that we covered when Argon debuted with cyber defenders first in line, to hundreds of additional partners.

What the Benchmarks Do and Do Not Show

It is worth separating two kinds of evidence here.

The first is Google's own reporting. The company cites 87 percent on the ERQA embodied reasoning benchmark, and 57 percent on RareBench and 52.2 percent on LabBench in private evaluations by Gamow Labs. On safety, it reports that attack success on the Gray Swan indirect prompt injection benchmark fell to 0.7 percent across 15 adaptive attempts. These are vendor figures and should be read as such.

The second is independent measurement. VentureBeat reports that Argon leads the Vals Index at 68.90 percent as of October 12, ahead of Claude Sonnet 5.5 at 67.04 percent and Claude Opus 5.5 at 66.97 percent. On the Artificial Analysis Intelligence Index it scores 53, with Opus 5.5 at 58 and Sonnet 5.5 at 56. The honest summary is that the frontier is closely contested, and that the right model depends on the workload.

One detail deserves care. Google describes a 1M unified input-output token limit, and VentureBeat notes that it is reached through a continuation protocol configured in the API, and that Google has not spelled out how it relates to the length of a single uninterrupted response. Developers planning very long outputs should test this directly.

What to Watch Next

The practical window is clear: the introductory rate runs for roughly seven weeks. That is enough time to rerun an existing evaluation suite, compare cost per completed task, and decide before the standard price arrives. The scientist access program is the part I find most encouraging, since it puts a frontier model in the hands of researchers who can test it on real problems. More model coverage is in our AI section.

Sources: Google blog, Introducing Gemini 4 Argon — October 13, 2026; Google AI for Developers changelog — October 13, 2026; VentureBeat — October 13, 2026.