Lotook
Gemini 4 Carbon (gemini-4-carbon) reached general availability on November 10, 2026, for paid Gemini API developers, enterprise customers, Google AI Ultra subscribers, and selected researchers. Through November 30, 2026, API access costs $2.00 per million input tokens and $10.00 per million output tokens, before doubling to $4.00 and $20.00 on December 1, 2026.
Google introduced Gemini 4 Carbon on October 28, 2026, as its next frontier reasoning model and opened broad access thirteen days later. Although Gemini 4 Carbon shares the exact same per-token API rate card as Gemini 4 Argon, the two models behave very differently on your monthly invoice and in production latency. This article covers who can access gemini-4-carbon right now, how the November promotional pricing works, how the model compares against competing frontier APIs, and when your engineering team should route prompts to Carbon instead of Argon.
What Is Gemini 4 Carbon and How Did It Roll Out?
Gemini 4 Carbon is Google’s frontier reasoning model in the Gemini 4 family, built for complex multi-step problems that require deeper exploration and verification per turn than Gemini 4 Argon. In the Gemini API, the model is exposed under the model ID gemini-4-carbon and accepts text, image, video, audio, and PDF inputs while producing text outputs.
Google rolled out Gemini 4 Carbon in two distinct phases during late October and November 2026:
- October 28, 2026 (Initial announcement and limited preview): Google announced Gemini 4 Carbon and restricted initial access to a subset of cyber defenders operating under the Fairwind Program. Google also noted on October 28 that the model was participating in the U.S. government’s voluntary pre-release testing.
- November 10, 2026 (General availability): Google made
gemini-4-carbongenerally available in the Gemini API changelog and opened access across developer tools, enterprise platforms, consumer subscriptions, and academic research applications.
If you have been tracking all of the Fall 2026 AI model releases, Gemini 4 Carbon arrives as Google’s heaviest reasoning option above Gemini 4 Argon, rather than a drop-in replacement for daily coding loops.
Who Can Use Gemini 4 Carbon Right Now?
Gemini 4 Carbon is available as of November 10, 2026, to developers on the Gemini API paid tier, enterprise teams, Google AI Ultra subscribers, eligible researchers, and Fairwind Program cyber defenders. According to Google’s official Gemini API pricing documentation, Gemini 4 Carbon is not available on the Gemini API free tier.
| User Group | Where Gemini 4 Carbon Is Available | Access Requirements and Notes |
|---|---|---|
| Developers | Gemini API (gemini-4-carbon) in Google AI Studio and Android Studio; Google Antigravity |
Requires a Gemini API paid tier billing account. Google AI Ultra subscribers can also test the model in Google Antigravity and AI Studio. |
| Enterprises | Gemini for business (API or model drop-down), Google Antigravity, CodeMender | Available via the API or the model drop-down in Gemini for business. |
| Everyday consumers | Gemini app and AI Mode in Google Search | Rolling out to Google AI Ultra subscribers. |
| Scientists and researchers | Google Antigravity (highest usage tier) | Researchers can apply for 6 months of free access to both Argon and Carbon, subject to eligibility, capacity, and selection. |
| Cyber defenders | Fairwind Program | Specialized access for defensive cybersecurity workflows. |
| Free-tier API users | Not available | Google explicitly excludes gemini-4-carbon from the unpaid Gemini API tier. |
Alongside the model rollout, Google announced that Google Antigravity is previewing a /swarm command. When triggered, /swarm converts a single prompt into a coordinated team of agents, allowing Gemini 4 Carbon to handle the deepest reasoning steps while lighter, faster models execute routine subtasks.
How Much Does the Gemini 4 Carbon API Cost?
Gemini 4 Carbon API pricing starts at an introductory rate of $2.00 per million input tokens and $10.00 per million output tokens through November 30, 2026, before doubling to $4.00 input and $20.00 output starting December 1, 2026. These rates match the introductory and standard schedule of Gemini 4 Argon on the paid Gemini API tier.
Every token processed by the model is billed. That count includes your visible prompt, the model’s visible reply, all internal thinking tokens (which Google bills at the output token rate), and internal tool system instructions or function declarations added when you enable grounding. On the paid tier, Google states that customer prompts and responses are not used to improve Google products.
| Pricing Tier (Paid API) | Through Nov 30, 2026 (per 1M tokens) | Starting Dec 1, 2026 (per 1M tokens) |
|---|---|---|
| Standard input | $2.00 | $4.00 |
| Standard output (includes thinking tokens) | $10.00 | $20.00 |
| Context caching (implicit only) | $0.10 | $0.20 |
| Batch / Flex input | $1.00 | $2.00 |
| Batch / Flex output | $5.00 | $10.00 |
| Batch / Flex caching | $0.05 | $0.10 |
| Priority input | $3.60 | $7.20 |
If your application connects gemini-4-carbon to live web results, Grounding with Google Search includes 5,000 free requests per month shared across Gemini 3 and newer models. After exhausting that monthly pool, Google charges $14.00 per 1,000 grounded search requests.
One caching rule matters for production architecture: Gemini 4 Carbon supports implicit context caching only. Explicit context caching is not supported on gemini-4-carbon, so teams looking for ways to reduce AI API costs should rely on consistent prompt prefixes to trigger implicit cache hits or move asynchronous workloads to the Batch / Flex tier, which cuts per-token prices by 50 percent both before and after December 1.
How Does Gemini 4 Carbon Pricing Compare to GPT-6 Astra and Claude 5?
Gemini 4 Carbon matches or undercuts competing frontier models on per-million-token list rates as of November 10, 2026, even after the December 1 price increase takes effect. For a broader look across providers, see our full AI API pricing comparison.
| Frontier Model | Standard Input (per 1M tokens) | Standard Output (per 1M tokens) | Pricing Notes (as of Nov 2026) |
|---|---|---|---|
| Gemini 4 Carbon | $2.00 (then $4.00) | $10.00 (then $20.00) | Introductory pricing ends Nov 30, 2026; Dec 1 rates double. |
| Gemini 4 Argon | $2.00 (then $4.00) | $10.00 (then $20.00) | Same per-token schedule as Carbon, but uses fewer thinking tokens per turn. |
| Claude Opus 5.5 | $4.00 | $20.00 | Matches Gemini 4 Carbon’s post-December 1 standard token rate. |
| Claude Fable 5.2 | $8.00 | $40.00 | Double the post-December 1 per-token rate of Carbon. |
| GPT-6 Astra | $10.00 | $50.00 | Five times Carbon’s introductory rate and 2.5x its December 1 rate. |
When comparing against Claude Fable 5.2 or GPT-6 Astra, keep in mind that list price per million tokens is only half of the cost equation. Total cost per task depends heavily on how many internal reasoning tokens each model burns before emitting a final answer.
What Technical Specs and Features Does gemini-4-carbon Support?
The gemini-4-carbon API endpoint provides a 1,048,576-token unified context window in which input tokens, thinking tokens, and output tokens all share the same capacity pool. For standard synchronous requests, the maximum output length is 65,536 tokens. When developers enable long decode for background execution, maximum output expands up to the full 1,048,576-token window minus the size of the input prompt.
Here is the capability breakdown from Google’s official model documentation as of November 10, 2026:
- Supported modalities: Text, image, video, audio, and PDF inputs; text output only.
- Thinking levels:
low,medium(the default setting), andhigh. - Supported tools and features: Function calling, structured outputs, code execution, computer use, file search, Google Search grounding, Google Maps grounding, URL context, and implicit context caching.
- Unsupported features: Explicit context caching, image generation, audio generation, and the Live API. (For creators who need dedicated visual generation rather than a text-only reasoning API, a multi-model studio such as AI Studio Video handles image and video generation workflows.)
What Benchmark and Safety Numbers Does Google Report?
Google reports first-place results for Gemini 4 Carbon on several coding, agentic, and cost-per-task benchmarks in its October 28 and November 10, 2026 posts. Every number in this section is vendor-reported by Google and has not been independently reproduced here.
Intelligence, Coding, and Agentic Claims
In its October 28, 2026 announcement, Google reported that Gemini 4 Carbon ranked #1 on the Artificial Analysis Intelligence Index v5.0 with a score of 59. In that same post, Google highlighted three cybersecurity and vulnerability benchmarks:
- SEC-bench Pro (finding vulnerabilities): 88.2%
- CWE-bench (patching vulnerabilities): 71.0%
- ExploitGym (exploitation): 54.8%
In the November 10, 2026 general availability announcement, Google added several cost-adjusted and agentic performance claims:
- CursorBench 4.0, Perplexity’s WANDR, and Zapier’s AutomationBench: Google states that Gemini 4 Carbon ranks #1 on all three benchmarks while operating at more than 2X lower cost per task compared to the next best external model.
- Multimodal Knowledge Agent (MKA) benchmark: Launched in partnership with Artificial Analysis, MKA places Gemini 4 Carbon at #2 overall while averaging more than 6X lower cost per task than the #1 model.
- Project Fortify (internal Google benchmark): Google reported 100% recall on vulnerability scanning and 95% precision on triage.
Safety and Red-Teaming Results
Google’s November 10 post compares Gemini 4 Carbon against Gemini 4 Argon on several internal and partner safety tests:
- Simulated cyberattack testing: Carbon blocked 99% of attack trajectories generated by an adaptive red-teaming agent, recording a per-attempt breach rate of 0.8% compared to 3.1% for Gemini 4 Argon.
- Gray Swan automated cyberattack benchmark: Query attack success dropped to 2.17% on Gemini 4 Carbon, down from 9.86% on Gemini 4 Argon.
- Agentic honesty (375-task benchmark): Gemini 4 Carbon passed 92% of tasks, up from 70% for Gemini 4 Argon.
- Prompt injection robustness: Google noted a 0% attack success rate on the Gray Swan indirect prompt injection (IPI) benchmark reported two weeks prior, while cautioning that prompt injection robustness remains an area for continued improvement.
Gemini 4 Carbon vs. Argon: Which Model Should You Use?
Choosing between Gemini 4 Carbon and Gemini 4 Argon comes down to how much reasoning depth your task requires and how sensitive your workflow is to per-turn latency and output token consumption. Both models share the same 1,048,576-token unified context window, the same set of built-in tools, and the exact same per-token API pricing ($2/$10 through November 30, 2026, and $4/$20 starting December 1, 2026).
However, identical price per token does not translate to identical cost per task. Gemini 4 Carbon is engineered to spend significantly more internal thinking tokens per turn exploring alternatives and verifying intermediate steps before returning an answer. Because Google bills every thinking token as an output token ($10 per million through November 30, then $20 per million), a single call to gemini-4-carbon will typically cost more and take longer to return than the exact same prompt sent to Argon.
Treating Carbon and Argon as cost-equivalent because their rate cards match is an easy mistake. If you are deciding how to choose an AI model inside the Gemini 4 lineup, follow Google’s own workload split:
- Use Gemini 4 Argon for the majority of software engineering and knowledge work: Argon uses fewer thinking tokens per turn and responds faster, making it the better fit for interactive coding assistants, day-to-day feature implementation, and fast production request loops.
- Reserve Gemini 4 Carbon for high-complexity reasoning bottlenecks: Google recommends Carbon for complex root-cause debugging, system and architectural design, dense multimodal synthesis across large documents or media, scientific proofs, and security auditing.
When you do deploy gemini-4-carbon, do not leave the thinking parameter unexamined. The model defaults to medium, and also supports low and high. Google’s engineering guidance is straightforward: run your evaluation suite across all three thinking levels (low, medium, and high), and lock in the lowest level that clears your quality bar. Dropping a routine pipeline from medium or high down to low trims thinking token generation directly, keeping both latency and December output bills under control.
FAQ
Is Gemini 4 Carbon available on the Gemini API free tier?
No, Gemini 4 Carbon is not available on the Gemini API free tier. Developers must enable a paid billing account in Google AI Studio or Android Studio to call the gemini-4-carbon model endpoint. Scientists and academic researchers can apply separately for 6 months of free high-tier access in Google Antigravity, subject to eligibility and capacity.
When does Gemini 4 Carbon API pricing increase?
Gemini 4 Carbon’s introductory API pricing runs through November 30, 2026, at $2.00 per million input tokens and $10.00 per million output tokens. Starting December 1, 2026, standard API rates double to $4.00 per million input tokens and $20.00 per million output tokens, with Batch / Flex and implicit caching tiers doubling on the same date.
Why does Gemini 4 Carbon cost more per task than Argon if token prices are identical?
Although Gemini 4 Carbon and Gemini 4 Argon share the exact same per-token rate card, Gemini 4 Carbon generates substantially more internal thinking tokens per turn to explore and verify solutions. Because the Gemini API bills every thinking token at the standard output token rate, Carbon consumes more billable output volume per prompt than Argon.
Does Gemini 4 Carbon support explicit context caching or the Live API?
No, gemini-4-carbon supports implicit context caching only and does not support explicit context caching. The model also lacks support for the Gemini Live API, image generation, and audio generation. It does support function calling, structured outputs, code execution, computer use, file search, URL context, and grounding with Google Search and Google Maps.
How big is the Gemini 4 Carbon context window and max output?
Gemini 4 Carbon features a unified context window of 1,048,576 tokens shared across input, internal thinking, and output tokens. By default, synchronous API calls support up to 65,536 output tokens, while enabling long decode for background execution allows outputs up to the full 1,048,576-token window minus the input token count.
Sources
- Google Blog: Gemini 4 Carbon Is Now Available (Nov 10, 2026)
- Google Blog: Introducing Gemini 4 Carbon (Oct 28, 2026)
- Google AI for Developers: Gemini 4 Carbon Model Documentation
- Google AI for Developers: What’s New in Gemini 4 Carbon
- Google AI for Developers: Gemini API Pricing
- Google AI for Developers: Gemini API Changelog
