Blog / Service updates

Gemini API Price Increase: Argon Doubles Dec 1, 3.8 Flash Jan 1

2026-10-18cdnping editors5 min read

Both of Gemini API's main models are getting more expensive, and the increase is a doubling: gemini-4-argon from December 1, 2026, and gemini-3.8-flash from January 1, 2027. A budget built on today's prices will fall short by year end. The numbers below come from the official Gemini API pricing page, checked as of October 17, 2026. Actual billing follows the official page.

#What are the old and new prices for each model?

The pricing page labels the current prices as limited-time rates. All figures are USD per million tokens.

gemini-4-argon (Standard)

Argon has no free tier, and caching is implicit only; explicit caching is not supported. Batch and Flex are priced the same: input $1.00 and output $5.00 now, then $2.00 / $10.00 from December 1. Google's Argon launch post (September 30) also calls $2 / $10 an introductory price, with $4 / $20 afterward. The post also says the model was first rolled out to trusted cybersecurity defenders, with broader access such as paid API customers coming later. It gives no date for that, so check the official sources for availability. You can also see our Gemini 4 Argon launch guide.

gemini-3.8-flash (Standard)

3.8 Flash has a free tier. Batch and Flex are priced the same: input $0.375 and output $1.875 now, then $0.75 / $3.75 from January 1. The free tier does not apply to these two modes.

#What does this mean for your budget?

Two main things. First, thinking tokens are billed at the output rate. A request with thinking enabled often produces far more output than the answer you see, so the output price doubling hurts more than the input price doubling. Second, Argon's increase arrives more than a month before Flash's, so if you plan to use Argon, your budget window is shorter.

Here is a hypothetical example: 1 million input tokens and 200,000 output tokens per day, no caching, over 30 days.

Plug in your own daily input and output volumes. The formula is: millions of input tokens × input price + millions of output tokens × output price.

#How can you bring costs down?

1. Move anything that can wait to Batch or Flex. On the official pricing page, these two modes cost half the Standard price, which suits offline evals, bulk summarization, and data cleanup. Check the official docs for latency and limits.
2. Make repeated prefixes hit the cache. Argon supports implicit caching only, so put unchanging content such as the system prompt and long documents at the very start of the request and the changing parts after it. The cache price is 5% of the input price (a 95% discount). If you use explicit caching on Flash, factor in the storage fee, which becomes $1.00 per million tokens per hour after the increase, so caches left sitting for a long time get more and more expensive.
3. Don't use Argon for simple tasks. Try Flash first for classification, formatting, and extraction, and only upgrade if quality falls short. For ideas, see how to do multi-model routing.
4. Control thinking usage. The output price includes thinking tokens, so for requests that don't need deep reasoning, look in the official docs for the thinking budget or thinking level parameter. Exact names are per the official docs.

#Will the billing cap block you?

The tiers listed on the official rate limits page are: Tier 1 only requires linking a billing account, with a billing cap of $250; Tier 2 requires $100 in cumulative payments and 3 days since the first payment, with a cap of $2,000; Tier 3 requires $1,000 in cumulative payments and 30 days, with a cap of $20,000 to over $100,000. The page does not say over what period the cap is measured, so check the official source. In the example above, Argon's monthly cost after the increase is about $240, which is already close to Tier 1's $250 cap. Teams with higher usage should plan their upgrade early.

#How do you get an alert before the increase?

Two ways. One is to subscribe to the RSS feed of this site's /changes page, which records line-by-line additions and removals on the Gemini pricing page, so you will see a diff when a price number changes. The other is to read how to find out right away when LLM API prices change, which covers all three channels: changelogs, model catalogs, and billing. To see whether Gemini is actually available right now, open the API status page /status/gemini.

If you want a fixed monthly budget paid in Chinese yuan (RMB), AGI Plan is one option: a monthly subscription billed in RMB, no overseas credit card needed, and four plans that differ only in quota. AGI Plan is a commercial partner of cdnping and is its commercial partner. Some models are available outside mainland China only, so check agiplan.dev/models for specifics. Billing varies a lot between routes, so compare them item by item with how to choose an API relay. In this context, a relay is a third-party gateway that resells model API access.

#Checklist for right now

• List every service that calls gemini-4-argon and recalculate its budget before December 1.
• List every service that calls gemini-3.8-flash and recalculate its budget before January 1, 2027.
• Find offline jobs that can move to Batch / Flex.
• Check whether your prompt prefix is stable enough to hit implicit caching.
• Confirm that your current tier's billing cap is high enough.
• Subscribe to the /changes RSS feed.

#Sources

• Gemini API pricing (official)
• Gemini API rate limits and usage tiers (official)
• Gemini 4 Argon launch post (Google official blog)