Same vendor · prices checked 2026-11-20
Neon vs Gemini 4 Carbon: which side of $60 a month
Carbon launches at $2.00 in and $10.00 out per million tokens, then moves to $4.00 and $20.00. At 10M input and 2M output a month that is $40.00, then $80.00 — while Neon sits at $60.00.
Neon vs Carbon: the rates
Both are Google models, so this comparison needs no cross-vendor translation — just the two rates and the period they apply in.
| Model | Input | Cached input | Output | Context |
|---|---|---|---|---|
| Gemini 4 Neon | $3.00 | $0.15 | $15.00 | 1.0M |
| Gemini 4 Carbon (introductory) | $2.00 | $0.10 | $10.00 | not published |
| Gemini 4 Carbon (after the introductory period) | $4.00 | $0.20 | $20.00 | not published |
Google's launch post states the introductory rates — $2.00 per million input tokens and $10.00 per million output tokens, the same introductory price as Gemini 4 Argon — and states in a footnote that $4.00 input and $20.00 output apply once the introductory period expires. The cached-input figures in the table follow from Google's own stated rule that cached input tokens are priced at 95% off the input token price: $2.00 less 95% is $0.10, $4.00 less 95% is $0.20. Those two are arithmetic on a published rule, not an estimate.
The launch post does not give the date the introductory period ends, so neither does this page.
What the price gap is, before and after
At the site's standard volume — 10,000,000 input and 2,000,000 output tokens per month, no cache hits:
| Model | Input cost | Output cost | Per month | Per year |
|---|---|---|---|---|
| Gemini 4 Neon | $30.00 | $30.00 | $60.00 | $720.00 |
| Gemini 4 Carbon (introductory) | $20.00 | $20.00 | $40.00 | $480.00 |
| Gemini 4 Carbon (after the introductory period) | $40.00 | $40.00 | $80.00 | $960.00 |
The order flips. During the introductory period Carbon costs $20.00 a month less than Neon (33.3% below the Neon bill); after it, Carbon costs $20.00 a month more (33.3% above). The annual figures move with it: $480.00 against $720.00 against $960.00.
That flip is the single most important thing on this page. A comparison written during the introductory period and read after it describes the opposite of today's prices.
What the extra $20 a month buys, as far as anything published says
This is the part where every comparison site becomes an opinion piece. Here is what is actually published, separated by who published it.
Google's own statements about Carbon (from the launch post, 2026-10-28): Carbon ranks on v5.0 of the Artificial Analysis Intelligence Index at 59; it scores 88.2% on SEC-bench Pro, 54.8% on ExploitGym and 71.0% on CWE-bench; it reduced successful cheating on Terminal-Bench 4.0 to 0.0%; and Google describes it as its strongest model for cybersecurity and long-horizon software engineering. All of these are vendor claims. Google's post also notes that third-party evaluations were run without guardrails, which is Google's own caveat about its numbers.
What this site has verified about Neon: an Artificial Analysis Intelligence Index value of 61 at the xhigh effort setting, third overall, and 1.0M tokens of context, both from sources listed on the benchmark page and checked 2026-11-18.
What we will not do: put 59 and 61 in one sentence as if they settled anything. Google's 59 is attributed to index version v5.0; the 61 for Neon comes from an Artificial Analysis article whose index version we did not verify. Different index versions are different measurements, so the two numbers are not a 2-point gap and this page does not present them as one. To make a real comparison someone would have to read both sources at the same version on the same day, and then still keep the price next to the score rather than on another page.
What nobody has published: an independent head-to-head cost-per-task figure for Neon against Carbon that we have verified. Cost per task needs task-level token counts from a controlled run; this site has not run one, so it prints no such number.
Context window: one figure, one blank
Gemini 4 Neon publishes a 1.0M token context window. The Carbon launch post we read does not state a context window, and its pricing table on Google Cloud lists rates per token, not tokens per window. Third-party listings we saw give figures that do not agree with each other, and a disagreement is not a source. The cell stays at not published until Google states it in a page we can cite.
What this comparison means for a bill, not for a ranking
Three shapes of workload, and where the flip lands:
- Input-heavy, no cache. Carbon's introductory $2.00 beats Neon's $3.00; after the period, Carbon's $4.00 loses to it. Same workload, opposite answer.
- Cache-heavy. Carbon's cached rate is $0.10 during the introductory period and $0.20 after, under Google's 95%-off rule; Neon publishes $0.15 flat. The advantage changes hands at the end of the period here too.
- Output-heavy. $10.00 against $15.00 during the period, $20.00 against $15.00 after it. The split is the same $5.00 either way, in opposite directions.
Enter your own numbers on the calculator: the Carbon rows are on the price matrix rather than in the calculator's three rows, so use the formula and the rates above.
Questions about this pairing
Should the comparison use the introductory price or the regular price?
Both, always, and labelled. A table that shows only the introductory price is accurate for a window it does not name; a table that shows only the regular price overstates today's bill. The two rows above exist so neither reader is misled.
Is the 95%-off cached-input rule the same as my provider's caching?
Not necessarily. Google states the discount on its own cached input tokens. What counts as a cache hit, how long it survives and what minimum applies are set by the provider's documentation, not by this page. The rate is what we publish; the mechanics are what decides how many tokens reach it.
Why not rank Carbon against Neon?
Because a rank needs a metric we would have to choose, and every choice here would be editorial. The figures that exist are in two separate columns with two separate owners on the benchmark page.
When will the “not published” cells change?
When the vendor publishes the figure, or when a source we can cite changes. Each row carries its own verification date, and the changelog records the day a row was added or a source was rechecked.
Next
- Gemini 4 Neon vs GPT-6.2 Sol: the cross-vendor pairing at the same volume.
- Gemini 4 Neon vs Claude Opus 5.5: the calculator with three models.
- Price matrix: every row, with the source and the date beside it.
Carbon rates from Google's launch post, published 2026-10-28, checked 2026-11-20. Neon rates verified 2026-11-18. ModelCost is not affiliated with Google or Anthropic.