Google's Gemini 4 Neon at $3 / $15 costs more than Gemini 4 Carbon until 1 December, then 25% less
Google released Gemini 4 Neon on 18 November 2026 at $3 per million input tokens and $15 per million output tokens, according to the Gemini API release notes. Google places Neon below Gemini 4 Carbon, its "strongest reasoning model", yet as of 19 November 2026 Neon costs 50% more per token than Carbon and Gemini 4 Argon, which the pricing page lists at $2 / $10 through 30 November 2026. For Carbon, that is an introductory price: Google's model guide says "Carbon's price is introductory through November 30, 2026". On 1 December Carbon and Argon double to $4 / $20, as I reported on 17 November 2026, and the order flips: Neon becomes 25% cheaper per token than either. For a team on Carbon or Argon, Neon has the lower per-token price only from that date.
Gemini 4 prices before and after 1 December 2026
Paid tier, US dollars per million tokens, input / output, from Google's Gemini API pricing page as of 19 November 2026. Output includes thinking tokens.
| Model and tier | Through 30 November 2026 | From 1 December 2026 |
|---|---|---|
| Gemini 4 Carbon and Argon, Standard | $2 / $10 | $4 / $20 |
| Gemini 4 Neon, Standard | $3 / $15 | $3 / $15 |
| Gemini 4 Hydrogen, Standard | $1.50 / $7.50 | $1.50 / $7.50 |
| Gemini 4 Carbon and Argon, Batch and Flex | $1 / $5 | $2 / $10 |
| Gemini 4 Neon, Batch and Flex | $1.50 / $7.50 | $1.50 / $7.50 |
| Gemini 4 Carbon and Argon, Priority | $3.60 / $18 | $7.20 / $36 |
| Gemini 4 Neon, Priority | $5.40 / $27 | $5.40 / $27 |
| Gemini 4 Carbon and Argon, cached input | $0.10 | $0.20 |
| Gemini 4 Neon, cached input | $0.15 | $0.15 |
Neon has one price for its whole 1-million-token context window and, in the words of the same model guide, "no introductory price". Like Carbon and Argon, it is not on the free tier. Every Neon figure is exactly twice Gemini 4 Hydrogen's, and Neon on Batch costs what Hydrogen costs on Standard.
Where $3 / $15 sits against Anthropic and OpenAI
Standard rates, US dollars per million tokens, as of 19 November 2026. OpenAI's figures are its short-context rates.
| Model | Input / output | Cached input | Source |
|---|---|---|---|
| OpenAI GPT-6.2 Astra | $8 / $40 | $0.40 | OpenAI pricing |
| Anthropic Claude Fable 5.2 | $8 / $40 | $0.20 | Anthropic pricing |
| Google Gemini 4 Carbon and Argon from 1 December 2026 | $4 / $20 | $0.20 | Gemini API pricing |
| Anthropic Claude Opus 5.5 | $4 / $20 | $0.20 | Anthropic pricing |
| Google Gemini 4 Neon | $3 / $15 | $0.15 | Gemini API pricing |
| Google Gemini 4 Carbon and Argon through 30 November 2026 | $2 / $10 | $0.10 | Gemini API pricing |
| Anthropic Claude Sonnet 5.5 | $2 / $10 | $0.10 | Anthropic pricing |
| OpenAI GPT-6.2 Sol | $2 / $10 | $0.10 | OpenAI pricing |
| OpenAI GPT-6.1 Sol | $2 / $10 | $0.10 | OpenAI pricing |
| Google Gemini 4 Hydrogen | $1.50 / $7.50 | $0.075 | Gemini API pricing |
None of the current models in this table shares Neon's price. On Anthropic's page, $3 / $15 belongs to the older Claude Sonnet 4.6 and Claude Sonnet 4.5. Neon is 25% below Claude Opus 5.5 and 50% above Claude Sonnet 5.5 and both GPT Sol models.
What a team on Carbon or Argon should test before 1 December
Google's What's new in Gemini 4 Neon page says that switching from Carbon, Argon or Hydrogen "requires no request schema changes". The model ID is the only required edit. One default is different: "Carbon and Argon default to medium and cap at high", while Neon defaults to high and adds xhigh. A straight swap therefore runs Neon at a deeper thinking level than the model it replaces, and thinking tokens are billed as output.
Google's launch post claims that Neon at its default high level has "a 32% lower cost per task than Carbon". It does not say which Carbon price it used. Because Neon's input and output prices stand in the same ratio to Carbon's, the arithmetic is simple, and it is mine: assuming the same input/output mix, at Carbon's $2 / $10 the claim requires Neon to use about 45% of Carbon's tokens per task; at $4 / $20 it requires about 91%.
In my view, the useful test before 1 December is to run Carbon at medium and Neon at medium and high on your own tasks, count tokens, and price both at December rates: $4 / $20 against $3 / $15. A test priced at this month's rates shows Neon 50% above Carbon per token; from 1 December it is 25% below. Work that can wait has a third option: Neon on Batch or Flex at $1.50 / $7.50 costs 25% less per token than Carbon on Batch from 1 December.
Why Google prices Neon there: my reading, and the other one
My reading is that Google priced Neon for the lineup that starts on 1 December, and built it as the landing place for Carbon and Argon customers when their bill doubles. The signs are on Google's own pages as of 19 November 2026. Neon has no introductory price, while Carbon's is due to end in under two weeks. The model guide now lists Carbon, Neon, Hydrogen and Gemini 3.5 Flash-Lite and does not mention Argon, which Google's pricing page still calls its "efficient Gemini 4 frontier reasoning model" with "lower token spend per turn compared to Gemini 4 Carbon". From 1 December that efficient model costs a third more per token than Neon. The model card says "Gemini 4 Neon is adapted from Gemini 4 Carbon", and some of Neon's frontier safety results are "carried forward from Argon".
The other explanation is that $3 / $15 reflects what Neon costs Google to serve, and the timing next to the Carbon increase is incidental. Google's own upgrade table points a different group at Neon: it is the recommended replacement for Gemini 3.1 Pro Preview, not for Carbon or Argon. On that view Neon fills the gap between Hydrogen and Carbon, and the 32% claim is the pitch.
Whichever reading holds, from 1 December Neon is the lowest-priced Gemini 4 model that Google describes as above Hydrogen, while Claude Sonnet 5.5 and the two GPT Sol models sit a third below it at $2 / $10.
My forecast: by 31 March 2027, Google's Gemini API docs, either the choosing-a-model page with its upgrade guidance or the deprecations page, will name gemini-4-neon as the recommended replacement or migration target for users of gemini-4-carbon or gemini-4-argon. As of 19 November 2026, the upgrade table points gemini-3.1-pro-preview to "gemini-4-neon (or gemini-4-carbon for your hardest problems)" and says nothing of Carbon or Argon users moving to Neon. If my reading is right, Google will say so once Carbon and Argon cost more per token than Neon. If on 31 March 2027 neither page names Neon as a replacement for Carbon or Argon, I was wrong.