Models
Gemini 4 Argon: nearly 25 trillion tokens in 72 hours, three times Gemini 3
Google says Gemini 4 Argon processed nearly 25 trillion tokens in the 72 hours after launch, about three times Gemini 3, and that it is short of capacity.
Google says Gemini 4 Argon, generally available since October 13, 2026, processed nearly 25 trillion tokens in the 72 hours after launch, about three times what Gemini 3 processed over the same period.
Where the figure comes from
- Sundar Pichai, Google's CEO, gave it in a Big Technology Podcast interview published on October 20: "The first 72 hours since we launched, it's almost up to 25 trillion tokens used, which is about three times what Gemini 3.0 had in the first 72 hours. And we are capacity constrained as we are serving it" (Big Technology).
- The official News from Google account repeated the figure (X), as did Koray Kavukcuoglu, SVP of Google DeepMind and Google's Chief AI Architect (X).
A Google figure, not verified. Google does not say how these tokens split between input and output, between paying customers and internal use, or between the Gemini app, Search and the API. Nor does it give Gemini 3's absolute volume (Aiholics). It is a usage measure, not a quality measure.
The volume is also inflated by the nature of the model: Argon accepts up to one million tokens of combined input and output through the API, and Google pushes it toward long agent tasks (Google). A coding-agent session consumes far more tokens than a chat reply.
Pricing: $2 / $10, but only until November 30
The $2 per million input tokens and $10 output price is an introductory price. According to the Gemini API pricing page (Google):
- Standard: $2 input and $10 output (thinking tokens included) through November 30, 2026, then $4 and $20 from December 1.
- Caching: $0.10 per million cached input tokens, then $0.20; only implicit caching is offered, not explicit caching.
- Batch and Flex: half price ($1 and $5, then $2 and $10).
- No free tier for Argon in the API.
These prices match our Gemini 4 Argon page and the model comparison; no data needed to change.
What Google says about coding use
- On the launch blog, Google says that internally, Argon in Antigravity increased agentic code submissions by more than 35% and cut code rollbacks by nearly a third (Google).
- Sundar Pichai, for his part, mentions a 30% increase in code changes (CLs) in four weeks and acknowledges that even in coding "there are pockets where we know we need to do better" (Big Technology).
These internal figures are not independently measured.
What it means for developers
- Budget: a coding agent running on Argon will cost twice as much per token from December 1. The cost calculator estimates the difference on your volumes.
- Capacity: Google says it is capacity constrained serving the model. Sundar Pichai adds that a customer asking today for compute in 2027 will find it "incredibly difficult" to get. Planning a fallback in case of usage limits is prudent.
- Access: Argon is available through the API (AI Studio paid tier and Gemini platform) and, for Google AI Ultra subscribers, in Antigravity. To move from Gemini CLI to Antigravity CLI, see our guide.
- Cheaper alternatives: the DeepSeek V4.1 Flash, MiniMax M3.1 Flash, Gemini 4 Argon comparison sets Argon against two "Flash" models that are far cheaper per token.
Sources
- Google: Gemini 4 Argon: Built to solve your hardest problems (Oct 13, 2026), verified 2026-10-21
- Google AI for Developers: Gemini API pricing (page updated Oct 19, 2026), verified 2026-10-21
- Big Technology: interview with Sundar Pichai (Oct 20, 2026), verified 2026-10-21
- News from Google on X (Oct 20, 2026), verified 2026-10-21
- Koray Kavukcuoglu on X (Oct 2026), verified 2026-10-21
- Aiholics: Gemini 4 Argon hit nearly 25T tokens in 72 hours (Oct 20, 2026), verified 2026-10-21
Key takeaway: 25 trillion tokens in 72 hours is a usage figure announced by Google; the $2/$10 price is an introductory rate that doubles on December 1, 2026.
Spotted an error in this article? Report it · Our corrections policy