# Gemini 4 Carbon Tops the Index. You Cannot Build On It Yet

> Gemini 4 Carbon ranks first at 59 but is gated to cyber defenders, with an introductory price that later doubles. What founders should build on instead.

- Canonical: https://zeroraised.com/blog/gemini-4-carbon-gated-frontier-model
- Published: 2026-10-29
- Author: ZeroRaised Team (https://zeroraised.com/blog/authors/zeroraised)
- Publisher: ZeroRaised (https://zeroraised.com/)
- Section: Growth
- Tags: artificial-intelligence, growth, saas, bootstrapping

Gemini 4 Carbon is the strongest model on the Artificial Analysis Intelligence Index v5.0, scoring 59, and most founders cannot call it. Google announced it on 28 Oct 2026 and is giving it first to a subset of cyber defenders ([Google](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/)). The useful question is what to build while the best model on the board is gated, discounted and certain to be overtaken.

**TL;DR**

- Carbon ranks first at 59, but Google is phasing access to cyber defenders and says it is not yet available to developers.
- The introductory price matches Argon at $2 input and $10 output per million tokens, then rises to $4 and $20.
- Argon launched two weeks earlier and scored 50 on the same index.
- Among released open weights the best score is 32, widening the frontier gap to 12 points.
- Cost per task and effort settings move your bill far more than rank.

**Disclosure:** ZeroRaised publishes this analysis and is a launch site for bootstrapped products. Facts come from the linked sources as checked on 29 Oct 2026; ZeroRaised has no commercial relationship with the companies named.

**A gated frontier model** is a top scoring model that a lab releases first to a restricted group, with wider developer access promised later. It is a model you can read about before you can call it.

## What did Google actually ship?

Google credits Koray Kavukcuoglu and puts Carbon first on the index at 59 ([Google](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/)). The evaluation numbers are security heavy: 88.2 percent on SEC-bench Pro for finding vulnerabilities, 54.8 percent on ExploitGym for exploitation and 71.0 percent on CWE-bench for patching ([Google](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/)). Reuters framed it as a cybersecurity focused model made available first to a subset of cyber defenders ([Reuters via Investing.com](https://www.investing.com/news/stock-market-news/google-launches-gemini-4-carbon-ai-model-with-cybersecurity-focus-4980426)).

Access is the catch. Google says it will make Carbon available to a subset of cyber defenders starting this week, then gradually to additional trusted partners through its Fairwind Program, before availability to developers, businesses and consumers later ([Google](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/)). The announcement states that Carbon is not yet available to developers ([Google](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/)).

A security launch wearing a model launch costume: the benchmarks, the audience and the access list all point the same way.

## Why did the leaderboard move again in two weeks?

Argon launched two weeks before Carbon ([Google](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/)). When Artificial Analysis published v5.0 on 27 Oct, Claude Opus 5.5 at max effort led on 58, ahead of GPT-6 Astra on 57 and GPT-6.1 Sol on 55, with Argon at high effort on 50 ([Artificial Analysis](https://artificialanalysis.ai/articles/artificial-analysis-intelligence-index-v5)). Carbon arrived the next day at 59.

The crown moved twice in a fortnight, the second time inside one family. Treat rank as a timestamp.

Sixty percent of the index now uses private test sets or held-out answers, and no single model leads all nine evaluations ([Artificial Analysis](https://artificialanalysis.ai/articles/artificial-analysis-intelligence-index-v5)). A model can be first overall and still lose the one task your product runs ten thousand times a month.

## What will Carbon cost when the introductory price ends?

The introductory price is the same as Argon: $2 per million input tokens and $10 per million output, with cached input 95 percent off the input price ([Google](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/)). After the introductory period it becomes $4 input and $20 output ([Google](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/)).

The pattern is normal. Argon launched on the same terms and lists at the same $4 and $20 ([Google](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/)). The discount is published, dated and easy to miss if you only read the launch post.

Model gross margin on the list price. If the feature only works at the launch rate, it does not work.

## Why does cost per task beat headline rank?

Cost per task varies by more than 13 times among the ten highest scoring models, from $0.96 for GPT-6.1 Terra at max effort to $13.04 for Claude Fable 5.1 at max effort ([Artificial Analysis](https://artificialanalysis.ai/articles/artificial-analysis-intelligence-index-v5)). In between, Opus 5.5 costs $10.04, GPT-6 Astra $6.24 and GPT-6.1 Sol $1.27 ([Artificial Analysis](https://artificialanalysis.ai/articles/artificial-analysis-intelligence-index-v5)). Sol scores within 4 points of Opus 5.5 at about an eighth of the cost per task ([Artificial Analysis](https://artificialanalysis.ai/articles/artificial-analysis-intelligence-index-v5)).

Effort can move one model as much as a swap. Opus 5.5 scores 35 at low effort and 58 at max, at ten times the cost per task, $0.98 against $10.04 ([Artificial Analysis](https://artificialanalysis.ai/articles/artificial-analysis-intelligence-index-v5)). Median output tokens per task also rose by two thirds, from 35,000 to 59,000 ([Artificial Analysis](https://artificialanalysis.ai/articles/artificial-analysis-intelligence-index-v5)).

Your bill is tasks times tokens times price. Rank appears nowhere in it.

## What happened to the open weights gap?

It widened from 3 points to 12 ([Artificial Analysis](https://artificialanalysis.ai/articles/artificial-analysis-intelligence-index-v5)). Kimi K3.1 Pro is the top open weights model at 46, but its weights had not been released; among released weights, DeepSeek V4.1 Flash 1017 and GLM-5.3 lead at 32, below Claude Haiku 5.5 on 36 ([Artificial Analysis](https://artificialanalysis.ai/articles/artificial-analysis-intelligence-index-v5)).

That 12 point gap is the number to sit with before self hosting. The [open weights versus closed models comparison](https://zeroraised.com/blog/open-weight-vs-closed-ai-models) covers when the trade pays.

## What this means for bootstrapped founders

**Build on what you can call today.** Carbon has no developer API. Ship on a model with a published endpoint and a price you can plan around. The [guide to choosing an LLM API](https://zeroraised.com/blog/how-to-choose-an-llm-api) is the checklist.

**Price on the list rate.** Carbon's introductory period ends on a date Google has not given, then doubles to $4 and $20 ([Google](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/)). Put the list rate in your model.

**Measure cost per task on your own traffic.** Index numbers are a warning, not an invoice. The [LLM cost reduction guide](https://zeroraised.com/blog/how-to-reduce-llm-costs) covers caching and routing, and the [AI feature guide](https://zeroraised.com/blog/how-to-add-ai-features-to-a-saas) walks through pricing a feature per task.

**Keep the integration thin.** Assume a switch within a quarter. A router turns leaderboard churn into a config change.

**Do not rebuild for a rank.** The index moved twice in two weeks. Your architecture should survive the next two moves.

## Where ZeroRaised fits

ZeroRaised is a launch and discovery site for products built without institutional venture capital or accelerator equity. If you have something people can try, [submit your product](https://zeroraised.com/submit) for a launch page that states what it does and what it costs. The [pricing page](https://zeroraised.com/pricing) explains the options. We do not promise traffic, rankings or a spot in any training set. If you want the next analysis, [join the ZeroRaised newsletter](https://zeroraised.com/newsletter).

## Sources

- [Google: Introducing Gemini 4 Carbon](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/)
- [Artificial Analysis: Intelligence Index v5](https://artificialanalysis.ai/articles/artificial-analysis-intelligence-index-v5)
- [Reuters via Investing.com: Google launches Gemini 4 Carbon](https://www.investing.com/news/stock-market-news/google-launches-gemini-4-carbon-ai-model-with-cybersecurity-focus-4980426)

## Frequently asked questions

### Can I use Gemini 4 Carbon in my product today?

No. Google says Carbon is being made available first to a subset of cyber defenders, then gradually to additional trusted partners through its Fairwind Program, before wider availability to developers, businesses and consumers. The company states that Carbon is not yet available to developers.

### How much does Gemini 4 Carbon cost?

Google lists an introductory price of $2 per million input tokens and $10 per million output tokens, the same as Gemini 4 Argon, with cached input at 95 percent off the input price. After the introductory period the price is $4 input and $20 output per million tokens.

### Is Gemini 4 Carbon really the best model?

It ranks first on Artificial Analysis Intelligence Index v5.0 at 59, which is a snapshot across nine evaluations. Artificial Analysis notes that no single model leads all nine, and 60 percent of the index now uses private test sets or held-out answers. Headline rank is not the same as best for your specific task.

### Should I wait for Carbon before building my AI feature?

No. Waiting on a gated model hands your schedule to a vendor rollout you cannot see. Build on a model you can call today, keep the integration thin, and switch when the model you want is generally available and priced on a list rate you can plan around.

### What is the open weights gap on the index?

Artificial Analysis says the gap between open weights and the frontier widened from 3 points to 12. Kimi K3.1 Pro is the top open weights model at 46, but its weights were not yet released; among released weights, DeepSeek V4.1 Flash 1017 and GLM-5.3 lead at 32, below Claude Haiku 5.5 at 36.