Gemini 4 Carbon is available to paying API developers, with launch pricing that doubles after November 30
Google has made Gemini 4 Carbon, its strongest model, available to all paying Gemini API developers. It costs $2 per million input tokens and $10 per million output until November 30, then $4 and $20. Specs, thinking levels and how to call it.
Updated November 20, 2026: Expanded with specs, thinking levels, a quickstart, cost examples and how Carbon compares with Google's other Gemini 4 models.
Google made Gemini 4 Carbon available to all paying developers and enterprises on November 10: through the Gemini API in Google AI Studio, and through Gemini for business. It is also rolling out to Google AI Ultra subscribers in Google Antigravity, the Gemini app and AI Mode in Search. Two weeks earlier, Google had released Carbon to a set of cybersecurity defenders in its Fairwind Program, which still takes applications on a rolling basis.
The price, and when it changes
Carbon launched at the same introductory price as Gemini 4 Argon: $2 per million input tokens and $10 per million output tokens on the Standard paid tier, where output includes thinking tokens. The pricing page also lists Batch, Flex and Priority tiers, priced differently, so check the tier you use. A footnote in Google's post says that after the introductory period ends on November 30, 2026, Carbon costs $4 per million input and $20 per million output. Context caching follows the same pattern, from $0.10 to $0.20 per million tokens. Only implicit caching is supported, so code that creates explicit caches will need changing. Carbon is not available on the free tier.
For a task that reads 200,000 tokens and writes 20,000 at Standard rates, that is $0.60 now and $1.20 from December 1. Google's pricing page lists the same before-and-after figures and dates for Argon, so on list price the two stay level after December 1 as well.
Search grounding is billed separately: 5,000 free search requests a month shared across Gemini 3 and newer models, then $14 per 1,000, with a daily cap. Google notes that with grounding, every processed token is billed, including thinking tokens and internal tool instructions. It says paid-tier data is not used to improve its products.
What Carbon is for
Google positions the models differently. Argon remains its recommended choice for most software engineering and knowledge work. Carbon is for the hardest problems: complex multi-step coding and debugging, architectural design, dense document synthesis, scientific proofs and security auditing. Between them, the docs list Gemini 4 Neon at $3/$15 for fast end-to-end task completion, and Gemini 4 Hydrogen at $1.50/$7.50, with a free tier, for everyday work. Google's developer docs say plainly that Carbon spends more thinking tokens per turn, which increases token use and latency, so the per-task bill can rise faster than the per-token price suggests.
Google says Carbon ranks first on CursorBench 4.0, Perplexity's WANDR and Zapier's AutomationBench, at less than half the cost per task of the next best outside model. Those are Google's own claims; its announcement gives no scores or method alongside them.
Specs
- Model code: gemini-4-carbon.
- Inputs: text, image, video, audio and PDF. Output: text.
- One unified context window of 1,048,576 tokens shared by input, thinking and output.
- Up to 65,536 output tokens per synchronous response; with long decode, up to the 1,048,576-token window minus your input.
- Thinking levels low, medium (the default) and high, all at the same per-token price.
How to call it
Google's examples use the Interactions API in the google-genai Python SDK. This one is adapted from its thinking-level example; the quickstart is the same call without generation_config. Replace the prompt with your own task and code:
from google import genai
client = genai.Client()
interaction = client.interactions.create(
model="gemini-4-carbon",
input="Analyze this retry logic for race conditions: <paste code>",
generation_config={"thinking_level": "medium"},
)
print(interaction.output_text)Two settings control cost. Google recommends running your evaluations at all three thinking levels and picking the lowest that meets your quality bar. And max_output_tokens caps thinking plus output together: the response stops when the two hit the limit, so set it high enough that thinking doesn't cut the answer short, and low enough to limit spend. Responses longer than 65,536 tokens need background=True with a higher max_output_tokens; the server then works in 64K hops, and you join the text of every model_output step.
Other places to use it
Google is previewing a /swarm command in Antigravity that turns one prompt into a coordinated team of agents you can track in real time. Carbon takes roles such as architectural design, coordination and auditing, with faster, lighter models working in parallel isolated workspaces and adversarial verifiers testing each change. Scientists can apply for a chance at free access to Argon and Carbon on Antigravity's highest usage tier for six months, subject to eligibility, capacity and selection. Carbon is also supported by the Antigravity agent in Gemini Managed Agents.
Related on claudefox
- Gemini 4 Hydrogen costs $1.50/$7.50, and it only gets much cheaper than Carbon on December 1
- Google's Gemini Decisions API scores up to 1,000 yes/no and multiple-choice questions in one call
- Product Engineering: 45 free agent skills for shipping a product with Claude Code and Codex
- Plannotator: review your coding agent's plans, diffs and messages before it acts
Our take
In our view, the introductory price makes November a cheap month to benchmark Carbon on your hardest tasks. Measure cost per finished task, not per token, and do it at more than one thinking level: a model built to think longer can erase its headline price advantage. After December 1, we expect most teams to use Carbon only where Argon, Neon or a lighter model clearly falls short.
Source: Gemini 4 Carbon is now available (Google), Gemini API pricing (Google AI for Developers), What's new in Gemini 4 Carbon (Google AI for Developers), Gemini 4 Carbon model page (Google AI for Developers)
This story first appeared in the claudefox daily briefing for November 12, 2026.