Gemini 4 Carbon is out
● Generally available 10 November — Unveiled 28 October with limited access through the Fairwind cybersecurity program; now open on the paid API tier.
● $2 in, $10 out per million — Introductory rate through 30 November. It becomes $4 and $20 after that.
● That is a quarter of Claude Fable 5.2 — Fable is $8 and $40. Even after the promotion ends, Carbon is half the price.
● 1M context, text, images, speech and video in — Proprietary and hosted. No downloadable weights.
Google moved Gemini 4 Carbon to general availability on 10 November 2026, a day after Anthropic launched Claude Fable 5.2. The timing is not subtle, and neither is the pricing.
The price, and what it does to the ladder
| Model |
Input / MTok |
Output / MTok |
Context |
| Gemini 4 Carbon to 30 Nov |
$2 |
$10 |
1M |
| Gemini 4 Carbon from 1 Dec |
$4 |
$20 |
1M |
| Claude Fable 5.2 |
$8 |
$40 |
1M |
| Claude Opus 5.5 |
$4 |
$20 |
1M |
| GPT-6 Astra |
$10 |
$50 |
1.05M |
Claude and GPT-6 Astra prices are from their vendors’ own model pages. Carbon’s pricing is as reported at launch — it is not yet listed on Google AI Studio’s public pricing page, which still shows only Gemini 4 Argon.
At the introductory rate Carbon costs one quarter of Claude Fable 5.2. After 1 December it is half. It also lands exactly on Claude Opus 5.5’s $4 / $20, which is the price Anthropic recommends most people start at.
Two dates to diary. Carbon’s introductory rate ends 30 November. Gemini 4 Argon’s price also doubles on 1 December, from $2 / $10 to $4 / $20. Any Gemini cost model built this month is built on numbers with weeks left.
Where it wins and where it does not
| Benchmark |
Gemini 4 Carbon |
Claude Fable 5.2 |
| Artificial Analysis Intelligence Index |
60 high |
65 max |
| Zapier AutomationBench |
53.88% at ~$2.12/task |
51.45% at ~$4.64/task |
| Terminal-Bench 4.0 |
65.2% |
61.6% |
| FrontierCode 2.0 |
40.6% |
49.1% |
| Terminal-Bench-Science 0.1 |
64.8% |
74.6% |
| FrontierSWE v2 |
61.2% |
69.9% |
| Cost per task Artificial Analysis |
~$7.76 |
~$26.54 |
Opus 5.5 scores 59 on the Intelligence Index and GPT-6 Astra 57. FrontierSWE figures are Anthropic-reported; the cyber numbers come from Google’s own October evaluation.
Carbon takes business workflow automation and terminal work. Fable keeps coding and scientific research by clear margins. The number that will decide most decisions is the last row: about $7.76 per task against about $26.54, roughly a third of the cost for a model that is close on general intelligence and ahead on some agentic work.
Where you can use it
- Google AI Studio on the paid API tier, and Android Studio.
- Google Antigravity, which Google calls an agent-first development platform, for Google AI Ultra subscribers.
- Google has not published a country-by-country rollout schedule.
- Inputs are text, images, speech and video; output is text. It is proprietary and hosted, with no self-hosting licence announced.
Frequently asked questions
How much does Gemini 4 Carbon cost?
$2 per million input tokens and $10 per million output through 30 November 2026, rising to $4 and $20 from 1 December. Cached input is reported at roughly $0.10 during the promotion based on Google’s previously announced 95% cached-input discount.
Is Gemini 4 Carbon cheaper than Claude Fable 5.2?
Yes, substantially. Fable 5.2 is $8 / $40 per million. Carbon is a quarter of that during the introductory period and half of it afterwards.
Which is actually better?
It depends on the work. Carbon leads on Zapier AutomationBench and Terminal-Bench 4.0. Fable 5.2 leads on FrontierCode, FrontierSWE and scientific research, and scores higher on the Artificial Analysis Intelligence Index at 65 against 60. Full head-to-head here.
When does the introductory price end?
30 November 2026. From 1 December it is $4 input and $20 output. Separately, Gemini 4 Argon’s price doubles on the same date.