Gemini 4 Argon Intro Ends Dec 1: Keep It, Switch Flash, or Move?

The Cloud post, dated October 13, 2026, says Argon launches at an introductory rate of $2 per million input tokens and $10 per million output tokens through November 30, 2026. After November 30, it says standard pricing of $4 per million input tokens and $20 per million output tokens applies. It also says cached input tokens are priced at 95 percent off. It does not print $0.10 or $0.20, and it does not print the words December 1.

The API price card prints both the dollars and the December 1 date. Standard paid tier, per 1 million tokens in USD:

Ninety-five percent off the introductory $2 input price is $0.10. Ninety-five percent off the later $4 input price is $0.20. That arithmetic matches the cache row. It is a check between the post’s discount and the card’s dollars, not a third source. Every Argon cache row on the card says implicit caching only, and that explicit caching is not supported. The Argon rows do not list an hourly storage price. Gemini 3.8 Flash’s cache row does. Do not invent a storage line for Argon.

Gemini 4 Argon is not available on the free tier. The card says every token the model processes is counted and billed, including thinking tokens and, when you use grounding, internal tool system instructions and function declarations. There is no separate thinking-token price on the card. Thinking tokens are inside the output price.

The paid tier’s own feature list says the Batch API is a 50 percent cost reduction, and that paid content is not used to improve Google’s products. Argon’s Batch input, output, and cache rows are half the standard rows, before and after December 1. The Flex rows list the same token prices as Batch. The Argon section does not define how Flex differs from Batch in latency, queueing, or eligibility. Use the row you actually call. Do not treat the matching dollars as a matching service.

Priority is higher than standard on every row. $3.60 is 1.8 times $2.00, and $7.20 is 1.8 times $4.00. The same 1.8 times ratio holds on the published output and cache figures. That ratio is arithmetic on the card, not a sentence Google printed. A Contributor tier is listed as a private preview, invite only, at $1.00 input, $5.00 output including thinking tokens, and $0.05 context caching, with “used to improve our products” set to yes. Those Contributor prices are not dated to December 1 on the row I read. They are not the default paid path, and they are not a migration you can choose without an invitation.

It applies to paid-tier Gemini API traffic for gemini-4-argon whose billing day is on or after December 1, 2026, at the standard, Batch, Flex, or Priority rates in the table. The introductory column applies through November 30, 2026. A call on November 30 is the $2/$10 standard card if that is the tier you used. A call on December 1 is the $4/$20 standard card. The Cloud post’s “after November 30” and the price card’s “starting December 1, 2026” describe the same boundary. This article uses December 1 because that is the date printed on the rate that will be charged.

It does not apply to the free tier, because Argon is not offered there. It does not apply to a subscription checkout. The earlier guide is the one that separated a Pro or Ultra monthly fee from a per-million API price. This guide does not reopen that subscription decision and does not retitle that page. It also does not apply to Gemini Enterprise Agent Platform until you read that platform’s own price page. The API notes say the prices can differ.

Grounding is a separate meter. On the standard and Priority Argon tables, Grounding with Google Search includes 5,000 free search requests per month, shared across Gemini 3 and newer models, then $14 per 1,000 requests. The same tables list a daily cap of 1,500 requests per day for that paid grounding. The notes say the Gemini Developer API default is 1,500 requests per day per project across paid tiers for Grounding with Google Search on Gemini 3 and newer models, resetting at midnight Pacific Time. A customer request may fan out into more than one search query, and each query is charged. Grounding with Google Maps is 5,000 free requests per month, shared the same way, then $14 per 1,000 search queries. The Batch and Flex Argon tables I read do not include those grounding rows. Do not assume a Batch call carries the standard grounding allowance.

The model blurb on the price card calls Argon a flagship reasoning model for complex software engineering, deep mathematical and scientific problem solving, autonomous multi-step agents, and long-decode generation up to 1 million tokens. That is Google’s description. It is not a benchmark measured for this article.

Keep gemini-4-argon when three things are true. You are already calling that model id on the paid Gemini API. The job is one you are unwilling to send to a different model id. You have recomputed the December rate with your own token counts and can pay it. The list rate doubles on standard input, output, and cache. A workload that is all output doubles. A workload that is all uncached input doubles. Cache does not stop the doubling, because $0.10 becomes $0.20.

Switch to Gemini 3.8 Flash, model id gemini-3.8-flash, when you want the card published for that model and you accept that it is a different model. The price page calls it the most intelligent Flash model, engineered for long-horizon software engineering, autonomous agents, and complex enterprise workflows. That sentence is also a product description, not a score. On the standard paid tier, Flash input is $0.75 per million tokens through December 31, 2026, and $1.50 starting January 1, 2027. Output, including thinking tokens, is $3.75 through December 31 and $7.50 starting January 1. Context caching is $0.075 then $0.15. Flash also lists storage at $0.50 per million tokens per hour through December 31, 2026, and $1.00 per million tokens per hour starting January 1, 2027. Standard Flash is free of charge on the free tier. Argon is not. A Flash migration that ignores the storage line can cost more than the token line if you hold a large cache for many hours.

Flash has its own later step, on January 1, 2027, not on December 1. Moving off Argon to dodge December 1 and then forgetting January 1 is a second bill. Batch Flash, through December 31, is $0.375 input and $1.875 output per million tokens on the rows under that model. This article uses Flash as the replacement because it is the next model on the same price page, not because a benchmark said it matches Argon.

Move the same Argon model id onto Batch when the work can use the Batch API and you want the half-price rows: $1.00/$5.00/$0.05 through November 30, then $2.00/$10.00/$0.10. Those December Batch rates equal today’s standard introductory rates. A job that can wait until Batch, and that does not need a grounding row the Batch table does not list, can keep the introductory standard bill level after December 1 by changing the API mode rather than the model. Flex lists the same token prices. Move to Flex only when you already know what Flex means in your account. This pricing section did not define it.

Move off the Gemini API price card entirely when the deployment is Gemini Enterprise Agent Platform or another surface whose price page you have not checked. The October 13 post’s availability sentence is not a promise that every surface shares the API card. Antigravity and Gemini for business are named as places Argon can be used. Their invoices are not reprinted here.

Stay off the Contributor preview unless you were invited. The data-use flag on that row is yes. The paid standard row’s data-use flag is no.

Budget thinking tokens as output tokens. If a request returns 200,000 visible output tokens and 800,000 thinking tokens, the output meter for this card is 1 million tokens, because the row is “output price (including thinking tokens).” The introductory charge for that million is $10. The December charge is $20. If your logs split the two counts, add them before you multiply. If your logs already show one output total, do not add thinking a second time.

Cache billing, as this article reads the two sources, replaces the input price for the cached tokens. Uncached input uses the input row. Cached input uses the context caching row. Introductory example, illustrative, not a customer bill: 2 million uncached input tokens, 8 million implicit cache-hit tokens, and 2 million output tokens including whatever thinking is inside that output figure. No grounding. Standard paid tier.

Formula: uncached millions × input rate + cached millions × cache rate + output millions × output rate.

Through November 30: 2 × $2.00 + 8 × $0.10 + 2 × $10.00 = $4.00 + $0.80 + $20.00 = $24.80.

Starting December 1: 2 × $4.00 + 8 × $0.20 + 2 × $20.00 = $8.00 + $1.60 + $40.00 = $49.60.

The same token mix without any cache, 10 million input and 2 million output: 10 × $2 + 2 × $10 = $40 through November 30, and 10 × $4 + 2 × $20 = $80 from December 1. Both rates doubled, so this particular mix doubled. A mix that added grounding would not double, because the $14 per 1,000 queries is not in the token step-up.

The same uncached 10 million / 2 million mix on Batch: 10 × $1 + 2 × $5 = $20 through November 30, and 10 × $2 + 2 × $10 = $40 from December 1. December Batch matches November standard on this mix.

The same cached mix on Gemini 3.8 Flash standard, using only the per-million cache price and setting storage hours to zero in the assumption, through December 31, 2026: 2 × $0.75 + 8 × $0.075 + 2 × $3.75 = $1.50 + $0.60 + $7.50 = $9.60. Storage is not zero unless you are not holding the cache. Eight million tokens stored for 24 hours at the published $0.50 per million tokens per hour is 8 × $0.50 × 24 = $96. That storage figure is larger than the $9.60 token figure. It is still illustrative. It uses the Flash storage rate the Argon rows do not list. If your Flash cache is implicit and you are not billed the hourly storage line, the $96 does not belong on the invoice. Check the bill before you subtract Argon on the strength of $9.60 alone.

Priority standard-equivalent of the uncached mix, introductory: 10 × $3.60 + 2 × $18 = $36 + $36 = $72, against $40 on standard. From December 1: 10 × $7.20 + 2 × $36 = $72 + $72 = $144. Pay Priority only when you are already on that tier for a reason the token card does not state.

1. List every project that calls gemini-4-argon, and note standard, Batch, Flex, or Priority.
2. Pull a week of input tokens, cached tokens, output tokens, and thinking tokens. Add thinking into output if the export split them.
3. Recompute with the formula above. Replace the sample millions with yours. Keep the published rates.
4. If the December standard total is acceptable, keep the model id and the standard card. Put the new rate in the budget on December 1, not on the October 13 announcement date.
5. If the job can wait, move it to Batch and budget the half-price row. Confirm you are not depending on a grounding line that the Batch table does not show.
6. If someone proposes Flex because the dollars match Batch, ask what Flex changes besides the price. The Argon price section did not say.
7. If the work can move to gemini-3.8-flash, recompute with the Flash rates and add storage hours × tokens × the hourly rate when that line is on the invoice. Put a reminder on January 1, 2027, when Flash’s own introductory column ends.
8. Leave Contributor, Enterprise Agent Platform, and any Vertex quota note out of the forecast until the page for that surface is the one you are reading. The pricing notes mention Vertex only as a higher default quota for search grounding, 1,000,000 search queries per day, not as Argon’s token price.
9. Keep the October 9 article for the subscription question it answered. Do not pay a Pro or Ultra seat from this page’s per-million rates, and do not use that old page as the December API rate.

Public assumption for every total in the previous section: the token counts in the paragraph, paid Gemini API, US dollars per million tokens, no taxes, no grounding queries unless a sentence adds them, and no enterprise discount. Change the counts and run the same formula. The result is not your invoice.

When does the Gemini 4 Argon introductory price end?

The Gemini API price card says $2.00 input and $10.00 output per million tokens through November 30, 2026, and $4.00 and $20.00 starting December 1, 2026. The October 13 Cloud post says the introductory rate runs through November 30 and that standard $4 and $20 pricing applies after that. Cache on the API card moves from $0.10 to $0.20 per million tokens on the same dates.

Are thinking tokens billed separately?

The paid Argon card labels the output row “output price (including thinking tokens).” The note above the table says every token processed is billed, including thinking tokens. There is not a second thinking rate on that card. Count thinking tokens inside the output millions, once.

Does cache stay at 95 percent off after December 1?

The Cloud post says cached input is priced at 95 percent off. It does not print a later cache dollar. The API card prints $0.10 per million tokens through November 30 and $0.20 starting December 1, implicit caching only. Those dollars are 5 percent of the $2 and $4 input rates. Explicit caching is not supported on the Argon rows.

Is Batch half price, and is Flex the same product?

Argon Batch rows are half the standard input, output, and cache prices, before and after December 1. The paid-tier feature list calls the Batch API a 50 percent cost reduction. Flex lists the same token prices as Batch. The Argon section does not define Flex beyond that table.

Should I switch to Gemini 3.8 Flash to avoid December 1?

Switch only if you are willing to call gemini-3.8-flash instead of gemini-4-argon. Through December 31, 2026, standard Flash is $0.75 input and $3.75 output per million tokens, then $1.50 and $7.50 on January 1, 2027. Flash also publishes an hourly cache storage price. This article does not claim the two models match on any task.