Gemini 4 Argon and Carbon double on December 1
Gemini 4 Argon and Carbon double on December 1, 2026, from $2 and $10 to $4 and $20 per million tokens. Hydrogen stays at $1.50 and $7.50.

Gemini 4 Argon and Carbon double on December 1, 2026, from $2 and $10 to $4 and $20 per million tokens. Hydrogen stays at $1.50 and $7.50.

On November 15, 2026, Google's Gemini API pricing page ↗ lists the same token prices for Gemini 4 Argon (gemini-4-argon) and Gemini 4 Carbon (gemini-4-carbon). ai.google.dev/gemini-api/docs/pricing redirects there. Both models are marked unavailable on the free tier. Output prices include thinking tokens. Context caching on these rows is implicit only. The picture above is ours. We did not run a billed job.
Standard, Batch, Flex, and Priority all step up on December 1, 2026. The figures below are per 1 million tokens.
| Tier | Through November 30 | From December 1 |
|---|---|---|
| Standard | $2 in, $10 out, $0.10 cache | $4 in, $20 out, $0.20 cache |
| Batch | $1 in, $5 out, $0.05 cache | $2 in, $10 out, $0.10 cache |
| Flex | $1 in, $5 out, $0.05 cache | $2 in, $10 out, $0.10 cache |
| Priority | $3.60 in, $18 out, $0.18 cache | $7.20 in, $36 out, $0.36 cache |
The Argon launch post ↗ (October 13) and the Carbon availability post ↗ (November 10) already set the post-November 30 rate at $4 and $20. The pricing page adds cache, Batch, Flex, and Priority.
Gemini 4 Hydrogen is the other Gemini 4 id on that page, and it does not step. Standard paid rates stay $1.50 input, $7.50 output, and $0.075 cached input. The standard free-tier cells say "Free of charge." Argon and Carbon do not. No Hydrogen row names December 1.
Argon also has an invite-only Contributor row at $1 input and $5 output, with no December date, and those prompts are marked as used to improve Google's products. That is not the public price.
This example is illustrative. It is not an invoice and not a workload we ran. Take 10 million input tokens and 2 million output tokens. Output includes thinking tokens. No search-grounding fees.
| Case | Calculation | Total |
|---|---|---|
| Standard, through November 30 | 10 × $2 + 2 × $10 | $40 |
| Standard, from December 1 | 10 × $4 + 2 × $20 | $80 |
| Batch or Flex, from December 1 | 10 × $2 + 2 × $10 | $40 |
If those 10 million input tokens are cache hits instead of fresh input, the cache line is 10 × $0.10 = $1.00 through November 30 and 10 × $0.20 = $2.00 from December 1. Output is still billed at the output rate in the table.
Laabidi ↗ scaled the same rates to 300,000 requests of 8,000 tokens in and 400 out. 300,000 × 8,000 = 2.4 billion input tokens, which is 2,400 million × $2 = $4,800. 300,000 × 400 = 120 million output tokens, which is 120 million × $10 = $1,200. Sum $6,000 at today's standard rate, $12,000 on December 1 standard, and $6,000 again on December 1 Batch. Flex uses the same token rates. His sentence names Batch only. The arithmetic matches the page. It is his scenario, not a job we ran. The page does not say Batch or Flex has Standard's latency.
Gemini 3.8 Flash standard paid rates are $0.75 input and $3.75 output through December 31, 2026, then $1.50 and $7.50 starting January 1, 2027. Cached input goes from $0.075 to $0.15 on that date. That doubling is real, and it is not December 1.
gemini-2.5-flash is still on the pricing page, at $0.30 per million for text, image, and video input and $2.50 output on the standard paid tier. We did not find a December 1 shutdown for it there. A claim that it retires that day is not in this post.
GitHub's November 13 changelog ↗ says Copilot will deprecate Gemini 4 Argon on December 14, 2026, across Chat, inline edits, ask and agent modes, and code completions, and names Gemini 4 Carbon as the suggested alternative. The same note deprecates Gemini 3.8 Flash in favor of Hydrogen. That is a picker change inside Copilot. It is not a second API price list.
"the first row is claude sonnet 5.5's price to the cent." — Snaki ↗ (@Snak1i), November 12. On Anthropic's pricing page ↗, Sonnet 5.5 is $2 input and $10 output per million, and Opus 5.5 is $4 and $20.
"The intro rate is a land grab. I price a model at its December rate before I build on it." — E1ev3n ↗ (@harden_369741), November 10
"Frontier model pricing now works like a cloud promo. Budget for the price after the promo." — Jerry Fragiskatos ↗ (@Aces1974), November 13
"Gemini 4 Argon doubles its price on December 1, and Google's Batch API cancels the increase exactly." — Laabidi ↗ (@Mohamed_abidii), November 14. The token math above matches. Flex matches it too.
"The free one is also the one that stays put" — Doriana ↗ (@astierun), November 15, on Hydrogen. The same post says 3.8 Flash doubles January 1. The pricing page puts that step on January 1, 2027.
Reddit was not reachable for this story.
Price December before you build on Argon or Carbon. A stack budgeted at Sonnet 5.5's $2 and $10 matches Opus 5.5's $4 and $20 on December 1, on the base rates. Fable 5.2 is $8 and $40. Snaki's line matches those base rates. Cache does not: Gemini's implicit cache goes from $0.10 to $0.20, Sonnet's cache hit is $0.10, and Opus's cache hit is $0.20, with separate cache-write prices on Claude.
Hydrogen is the Gemini 4 model that stays put. $1.50 and $7.50, a free tier on the standard tab, no December step. Our Hydrogen note records Google's Artificial Analysis figure of 51 against Argon at 50. That number is Google's. Carbon's note already prints the December 1 table, and Google says Carbon uses more tokens than Argon. A lower price per million is not a lower bill per finished task.
Batch and Flex match today's standard token price after December 1. They are still Batch and Flex. If the job cannot use that tier, set the alert at $4 and $20.
Copilot users on Argon have a second date. On December 14, GitHub's changelog removes Argon from Copilot and names Carbon as the alternative. Carbon will already be on the higher API list price. The changelog is about the picker, not a claim that Copilot resale equals that list. Seat comparisons are in ChatGPT vs Claude vs Gemini and the best AI for coding guide.