Gemini 4 Carbon Leads AI Index at 59, Priced at Half of Opus

Google said on 28 October 2026 that Gemini 4 Carbon, “our next frontier model”, “ranks #1 on the recently updated v5.0 of Artificial Analysis Intelligence Index at 59”. In the benchmark table Google published with the launch, Claude Opus 5.5 scores 58 and GPT-6 Astra 57.

Google says Carbon “will launch at the same introductory price as Argon at $2 per million input tokens and $10 per million output tokens”. Anthropic’s pricing page lists Opus 5.5 at $4 and $20, so by our arithmetic Carbon is priced at half. For now, Google says it will make Carbon “available to a subset of cyber defenders” this week.

Short explainer · Gemini 4 Carbon brief

A vertical news Short narrated from the published headlines and opening lines of this piece and four related FourWeekMBA pieces on AI models, prices and chips.

Business Pill · Model routing

An evergreen Business Engineer explainer on model routing: sending each request to the cheapest model that can handle it. It explains the general idea only and does not discuss Gemini 4 Carbon.

What Google’s Table Shows

Google’s table compares Carbon with Gemini 4 Argon, GPT-6 Astra, Claude Opus 5.5 and Claude Fable 5.1 on 17 benchmarks. Carbon has the top score on 15 of them.

The two it does not lead are FrontierSWE 2, where GPT-6 Astra scores 65.5% against Carbon’s 61.2%, and the offline subset of OSWorld-2.1, where Opus 5.5 scores 87.2% against 81.5%.

On GraphWalks, a long-context test between 256,000 and 1 million tokens, Carbon scores 96.2% against 84.2% for Argon, 71.8% for GPT-6 Astra and 66.8% for Opus 5.5. On Vending-Bench 2, its simulated bank balance is $16,484, against $15,515 for GPT-6 Astra and $9,235 for Opus 5.5.

Google adds that Carbon “jumps +89 elo points to land at #4 (1769) on WebDev Arena”, so on that public leaderboard it is not first.

What Was Posted on X

An X user posted that Google’s Chrome team says Gemini 4 Carbon performed over 10 times better than another lab’s large frontier model at generating high-quality security patches, and that Google reports over 2 times higher secure-code success than Argon, nearly 80% fewer high-severity misalignment alerts and zero successful attacks on its evaluated prompt-injection benchmark.

TestingCatalog posted that Gemini 4 Carbon (High) took first place on Text Arena with 1528 points, ranked second in Agent Arena behind Opus 5.5 and fourth in Web Dev Arena behind Claude Opus 5.5, GPT-6 Astra and Sonnet 5.5.

Wall St Engine posted that Elon Musk called Google’s new Gemini 4 Carbon AI model “impressive.”

Elon Musk’s post reads, in full: “Impressive”.

Demis Hassabis posted that Gemini 4 Carbon is state of the art across key benchmarks and combines frontier capabilities with deep safety and security research.

alphaXiv posted that Gemini 4 Carbon is now state of the art on Terminal-Bench-Science at 64.8% versus GPT-6 Astra’s 63.3%, and that Carbon won’t be generally available for another two weeks.

The Cyber Numbers

Google describes Carbon as having “state-of-the-art cybersecurity and long horizon capabilities”. Its table puts Carbon at 88.2% on SEC-bench Pro, which tests finding real vulnerabilities, against 85.4% for GPT-6 Astra and 83.4% for Argon.

On ExploitGym, an exploitation benchmark, Carbon scores 54.8% against 42.4% for GPT-6 Astra and 34.5% for Opus 5.5. On CWE-bench, for patching, it scores 71.0% against 68.0% for Astra and 67.0% for Opus 5.5. Google says these were “evaluated without guardrails”.

Google also writes that “the Chrome team has reported that Carbon performed more than 10X better than another lab’s large frontier model when generating high quality patches”. It does not name that model.

Why the Rollout Is Gated

Google says it is “actively engaged in the U.S. government’s voluntary process for pre-release model testing”. Beyond the first cyber defenders, it says it “will gradually phase access to additional trusted partners through our Fairwind Program”.

Developers, businesses and consumers come later. Google says it is “continuing to gather feedback and iterate on guardrails before making Carbon available to developers, businesses, and consumers via our product surfaces as soon as possible”. Sundar Pichai wrote on X that this comes “ahead of wider availability soon, so hang tight!”

On safety, Google says Carbon “reduced successful cheating on Terminal-Bench 4.0 to 0.0%”, and that on Gray Swan’s prompt-injection benchmark it “achieves 0.0%”, against 0.7% for Argon. It also says that in live coding with its own teams, “high-severity misalignment alerts dropped by almost 80%”.

The Price Gap

Per million tokens, the official price pages list Opus 5.5 at $4 for input and $20 for output, and GPT-6 Astra at $10 and $50. OpenAI’s GPT-6.1 Sol is listed at $2 and $10, the same as Carbon’s introductory price.

By our arithmetic, Carbon’s introductory price is half of Opus 5.5’s and one-fifth of GPT-6 Astra’s, with Carbon one point and two points ahead of them on the Index score Google cites.

Connecting the Coverage

This morning Artificial Analysis’s new Index had Opus 5.5 on top: we reported Opus 5.5 tops AI Index at 58; GPT-6.1 Sol costs 8x less. Google now cites Carbon one point higher.

The price side follows our angle that AI prices fell 35%, yet Anthropic’s existing sales grew 9x. The chips under Gemini are in demand too: we reported that OpenAI plans to buy Google’s AI chips. On 27 October, Google DeepMind shipped Genie 4, its first real-time world model with sound.

Alphabet reports third-quarter results after the US market close today.

The Structural Read

As we read it, the gated rollout is part of the story: Google says it is engaged in the U.S. government’s voluntary process for pre-release model testing and will phase access through its Fairwind Program.

By our arithmetic, Carbon’s introductory price is half of Opus 5.5’s list price and one-fifth of GPT-6 Astra’s, for an Index score one and two points higher in Google’s table.

Google’s own table shows two benchmarks where Carbon is not first: FrontierSWE 2, where GPT-6 Astra scores 65.5%, and OSWorld-2.1, where Opus 5.5 scores 87.2%.

Three Implications

FOR ANTHROPIC Google’s table puts Claude Opus 5.5 one point behind Carbon on the Index and ahead of it on OSWorld-2.1, and Anthropic’s pricing page lists Opus 5.5 at $4 and $20 per million tokens.

FOR OPENAI OpenAI’s price page lists GPT-6.1 Sol at the same $2 and $10 as Carbon’s introductory price, and in Google’s table GPT-6 Astra leads FrontierSWE 2 at 65.5%.

FOR SECURITY TEAMS Google says Carbon goes first to a subset of cyber defenders this week, and its table puts Carbon at 88.2% on SEC-bench Pro, evaluated without guardrails.

The Business Engineer Lens

This story maps onto the Business Engineer framework The Open vs Closed Meta-Framework.

The framework starts from the line “Every technology stack has layers.” It sorts those layers into scarce and abundant ones and argues for keeping the scarce layer proprietary while opening the abundant one.

Gemini 4 Carbon is a closed model that Google is releasing first to a subset of cyber defenders, at an introductory price it says matches Argon’s. The framework was published in March 2026 and does not discuss Gemini 4 Carbon.

What Is Not Established

The benchmark scores are Google’s, from the table in its launch post, and the Index score of 59 is as Google cites it. Google links a methodology page for the evaluations; this piece relies on Google’s table for every comparison.

The price is described as introductory, and Carbon is not yet available to developers, so no public price page lists it. We did not contact Google, Anthropic or OpenAI.

The Bottom Line

Google says Gemini 4 Carbon scores 59 on Artificial Analysis’s Intelligence Index v5, one point above Claude Opus 5.5 in Google’s table, at an introductory price of $2 and $10 per million tokens, half of Opus 5.5’s list price. It goes first to a subset of cyber defenders, with wider availability promised soon.