Gemini 4 Argon API Pricing: $2/$10 Until Nov 30, Then It Doubles

Google's newest flagship landed on October 13, 2026, and the price tag is the headline: $2 per million input tokens and $10 per million output tokens — but only until November 30. From December 1, both rates double to $4 and $20. That makes Argon the first frontier-class model I've seen with a hard expiry date on its introductory pricing, and it changes the math for anyone deciding between Argon, GPT-6 Astra, and GPT-6.1 Sol right now.

All figures below come from Google's official Gemini API pricing page, which I read on October 15, 2026, and from our own pricing data layer for OpenAI's side (verified October 10, 2026). Prices verified October 15, 2026 — always confirm on the official pricing page before you commit a budget.

The direct answer

Gemini 4 Argon (API model ID gemini-4-argon) is available now through the paid Gemini API, Google AI Studio, and for Google AI Pro/Ultra subscribers in the Gemini app. Standard tier: $2.00 input / $10.00 output per million tokens through November 30, 2026; $4.00 / $20.00 from December 1. Output pricing includes thinking tokens — reasoning traces bill at the output rate, same as other Gemini models. Context caching costs $0.10 per million cached tokens during the intro period ($0.20 after), and it's implicit caching only: Google decides what gets cached, you can't explicitly pin a cache the way you can with OpenAI's cache writes. And one thing the launch coverage buries: Argon is not available on the API's free tier. You pay to experiment.

The full rate card

Google lists four processing tiers, each with an intro rate and a post–Nov 30 rate. Prices per million tokens, paid tier, USD:

Batch and Flex run at exactly half the Standard rate — a real 50% off for work that can tolerate delayed or lower-priority processing. Priority costs 1.8× Standard, the inverse of the patience discount. The doubling on December 1 applies to every tier, so nothing on this card escapes it.

What the doubling actually costs you

Let's run a realistic coding-agent task: 100,000 input tokens of code context plus 20,000 output tokens (including the model's thinking trace, which Google bills at the output rate).

• Argon, intro price: (0.1M × $2.00) + (0.02M × $10.00) = $0.20 + $0.20 = $0.40 per task
• Argon, from Dec 1: (0.1M × $4.00) + (0.02M × $20.00) = $0.40 + $0.40 = $0.80 per task
• Argon, Batch, intro price: (0.1M × $1.00) + (0.02M × $5.00) = $0.20 per task

At 1,000 tasks a month, the intro window saves you $400 a month — $4,800 a year if the rate held, which it won't. The practical takeaway is about timing, not arbitrage: prototype and benchmark on Argon now, at half the price you'll pay for the same tokens in December. If your workload goes to production in January, budget at the $4/$20 rates, not the ones on the launch blog.

Against the field

Here's where the price war gets interesting. Token-for-token, Argon's intro pricing is exactly GPT-6.1 Sol's pricing: $2/$10, identical. Run the task above on 6.1 Sol (from our data layer, verified October 10) and you get the same $0.40. Google's "undercut the competition" headline is really a 47-day window where a new flagship rents at mid-tier prices. After December 1, Argon at $4/$20 costs exactly twice what 6.1 Sol costs — and lands on the same line as Anthropic's Claude Opus 5.5 ($4/$20).

Against GPT-6 Astra ($10 input / $50 output), the contrast is brutal: the same 100K-in/20K-out task costs $2.00 on Astra — 5× Argon's intro price, and still 2.5× Argon's December price. Sol 6.1 keeps its speed tiers (Fast at 2×, and the Ultrafast mode I covered last week), but on pure token price, Google is telling you that frontier-class reasoning doesn't have to cost frontier-class money — at least for the next six weeks.

The fine print that matters

Thinking tokens bill at output rates. Argon is a reasoning model, and Google's page is explicit: output price includes thinking tokens. If your task triggers heavy reasoning traces, the effective output side is larger than your visible answer — budget for it.

Implicit caching only. Cached input is cheap ($0.10/M intro), but Google decides what gets cached — you can't explicitly pin a system prompt like you can with OpenAI's cache writes. On 6.1 Sol, cached reads are also $0.10/M, but you control them. For workloads built around stable long contexts, that's a real operational difference.

No free tier. You need a paid Gemini API plan to touch Argon at all. That raises the barrier for quick experiments — no free taste before the $2/$10 meter starts.

The bottom line

Gemini 4 Argon is a genuine pricing event: a frontier model at mid-tier prices, with a printed expiry date. If your workload is the kind Argon is built for — agentic coding, multi-step planning, long-decode generation — the next six weeks are the cheapest they'll ever be. Run the shootout now: Argon vs 6.1 Sol at identical intro prices is a fair fight, and it's the one worth running before December 1 makes it unfair. Just don't build your 2027 budget on November's prices. Our GPT-6 vs Gemini comparison has the standing rate card, and the calculator will do the per-workload arithmetic with your own token counts.

Sources: Google Gemini API pricing (read October 15, 2026); Reuters on the Argon launch; OpenAI-side figures from our pricing data layer, verified against the official OpenAI pricing page October 10, 2026. Prices verified October 15, 2026.