GOOGLE'S MOST CAPABLEGEMINI 41M CONTEXTSAME RATES AS ARGONHALF PRICE TO NOV 30
Gemini 4 Carbon API Pricing
Gemini 4 Carbon is the model Google's price list calls its most capable reasoning model, announced on October 28, 2026 for a subset of cyber defenders and generally available on the Gemini API since November 10, 2026. It costs exactly what Gemini 4 Argon costs: an introductory $2.00/M input, $10.00/M output and $0.10/M cached input through November 30, 2026, then $4.00 / $20.00 / $0.20 from December 1. The difference is in the bill, not the rate card: Google says Carbon spends more thinking tokens per turn, and those bill at the output rate. Checked by hand against Google's Gemini API pricing page.
Input - per 1M tokens
$2.00/M
$4.00 from Dec 1, 2026 -50%
Output - per 1M tokens
$10.00/M
$20.00 from Dec 1, 2026 -50%
Cached input - implicit only
$0.10/M
$0.20 from Dec 1, 2026 -95%
Effective - agentic blend
$1.21/M
92/8 split - 82% cache
§ 01 / TERMINAL
Gemini 4 Carbon API cost calculator.
Live calculator on Google's introductory rate, which runs through November 30, 2026. Every figure here doubles on December 1. Carbon thinks longer than Argon on the same task, so for a like-for-like estimate give it a larger output share than you would give Argon.
$ /mo
Workload split
Prompt cache hit rate
Tokens you can process
—
Words equivalent (English)
—
Effective rate
—
§ 02 / SCENARIOS
Gemini 4 Carbon cost on real workloads.
LONG-HORIZON CODING
Codebase migration step
$0.288/step
CYBER DEFENSE
Find-and-patch vulnerability run
$0.216/run
KNOWLEDGE WORK
Financial research brief
$0.115/brief
LONG OUTPUT
Single-pass module rewrite
$2.03/rewrite
Estimate · gemini-tokenizer-estimate · ≈3.85 chars/token Auto-counts as you type
This is a chars-per-token approximation, not a real tokenizer. Actual tokens vary by language, code density, and tool-call overhead — counts are typically ±10–20% off for English prose, more for code or non-Latin scripts. For exact billing, use the vendor's official tokenizer.
Characters 719
Words 115
Tokens (estimated) 187 tokens
Cost as input · uncached $0.00037 USD
Cost as output · uncached $0.00187 USD
Cost as cached input $0.00002 USD
| Model | Input /M | Output /M | Effective blended | Context | Best for |
|---|---|---|---|---|---|
| Gemini 4 Carbon Current | $2.00 cache $0.10 | $10.00 | $1.21 agentic 92/8 | 1M | Google's most capable reasoning model |
| Gemini 4 Argon | $2.00 cache $0.10 | $10.00 | $1.21 same · promo | 1M | Same rates, fewer thinking tokens per turn |
| Gemini 3.8 Flash | $0.75 cache $0.075 | $3.75 | $0.481 −2.5× · promo | 1M | Newest Gemini Flash - high volume at a fraction of the rate |
| Gemini 3.1 Pro Preview | $2.00 cache $0.20 | $12.00 | $1.44 +20% | 1M | Google's Pro tier, still in preview |
| GPT-6 Astra | $10.00 cache $1.00 | $50.00 | $6.82 +5.7× | 1.05M | OpenAI's top of stack |
| GPT-6.1 Sol | $2.00 cache $0.10 | $10.00 | $1.29 +7% | 1.05M | Balanced GPT-6 for coding and agents |
| Claude Fable 5.2 | $8.00 cache $0.20 | $40.00 | $5.01 +4.1× | 1M | Anthropic's top tier |
| Muse Wonder | $4.00 cache $0.20 | $20.00 | $2.41 +2.0× | 1M | Meta's most capable model, in preview |
Compare siblings
Gemini 4 Carbon has published scores on 9 benchmarks we track, and leads 4 of them. All of them are the lab's own reports; each row links to the page that published it.
| Benchmark | Score | Rank | Leader | Source |
|---|---|---|---|---|
| Terminal-Bench 4.0 Agents and computer use | 65.2% Artificial Analysis harness | #2 of 34 | Kimi K3.1 Pro · 65.7% | Vendor-reported Google DeepMind |
| ExploitGym Cybersecurity | 54.8% via the Fairwind Program | #1 of 9 | Gemini 4 Carbon leads | Vendor-reported Google DeepMind |
| SEC-bench Pro Cybersecurity | 88.2% via the Fairwind Program | #1 of 8 | Gemini 4 Carbon leads | Vendor-reported Google DeepMind |
| AutomationBench Agents and computer use · Main set (as reported) | 53.9% as reported | #3 of 22 | MiniMax M3.1 Flash · 62.8% | Vendor-reported Google DeepMind |
| Terminal-Bench-Science AI research | 64.8% Artificial Analysis harness | #4 of 12 | Claude Mythos 5.2 · 75.3% | Vendor-reported Google DeepMind |
| Chartography Vision and documents · Tool use not stated | 72.5% tool use not stated | #1 of 2 | Gemini 4 Carbon leads | Vendor-reported Google DeepMind |
| OSWorld 2 Agents and computer use · 2.1, offline subset | 81.5% offline subset | #3 of 5 | Claude Fable 5.2 · 89.3% | Vendor-reported Google DeepMind |
| FrontierSWE Coding · FrontierSWE v2 | 61.2% as reported | #2 of 3 | Claude Fable 5.2 · 69.9% | Vendor-reported Google DeepMind |
| Agents' Last Exam Agents and computer use · ALE v1.1 score | 77.4% v1.1, score | #1 of 1 | Gemini 4 Carbon leads | Vendor-reported Google DeepMind |
Gemini 4 Carbon API pricing FAQ.
What Gemini 4 Carbon costs now and after November 30, why it bills more than Argon at the same rates, and how it compares with the other frontier models.
Q · 01 How much does Gemini 4 Carbon cost? +
Google lists
$2.00/M input, $10.00/M output (thinking tokens included) and $0.10/M cached input through November 30, 2026. Under this site's 92/8 agentic blend at an 82% cache-hit rate that is $1.21/M effective. There is no long-context tier: one rate across the whole window.Q · 02 What changes on December 1, 2026? +
Every token rate doubles: input to
$4.00, output to $20.00 and cached input to $0.20 per million. Batch, Flex and Priority double on the same date. It is the same schedule Gemini 4 Argon is on. This page stores the rate in force today and will record the change when it happens; until then it shows both.Q · 03 Carbon and Argon cost the same per token. Which is cheaper to run? +
Usually Argon. The two share a window, tools and every per-token rate, but Google's migration guide says Carbon spends more thinking tokens per turn to explore and check its work. Thinking tokens bill at the output rate,
$10.00/M, so the same request produces a larger bill and takes longer on Carbon. Google suggests Carbon for root-cause debugging, architecture, scientific proofs and security audits, and Argon for interactive coding and fast production loops.Q · 04 Can I use Gemini 4 Carbon through the API today? +
Yes, since November 10, 2026, as
gemini-4-carbon on the paid tier of the Gemini API and in Google AI Studio. Google's pricing page says it is not available on the Free tier. From the October 28 announcement until then, only a subset of cyber defenders in Google's Fairwind Program had access.Q · 05 What is the context window? +
One unified window of
1,048,576 tokens shared by input, thinking and output. A synchronous call returns up to 65,536 output tokens by default; with long decode a single response can run up to 1M tokens. All of it bills at the output rate, the highest on the card, and Carbon thinks longer than Argon, so cap max_output_tokens on long jobs.Q · 06 Are there Batch, Flex or Priority rates? +
Yes, all three. Batch and Flex are half of Standard:
$1.00/M input and $5.00/M output. Priority is 1.8x Standard: $3.60/M input and $18.00/M output through November 30, 2026. Caching is implicit only: Google does not support explicit context caching on Carbon, so there is no cache-storage charge.Q · 07 How does Carbon compare on price with GPT-6 Astra, Claude Fable 5.2 and Muse Wonder? +
Under this site's blend Carbon works out to
$1.21/M, against $6.82/M for GPT-6 Astra, $5.01/M for Claude Fable 5.2 and $2.41/M for Muse Wonder. The blend is per token; a model that writes more thinking tokens for the same answer narrows or widens the gap on a real bill.Q · 08 What does Google say Carbon is good at? +
Google's launch figures: first on the Artificial Analysis Intelligence Index v5.0 at 59, 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation and 71.0% on CWE-bench for patching, and fourth on WebDev Arena (1768). These are Google's own numbers from the announcement, not independent runs; the scores this site records are in the benchmark table on this page.