🔍 Read the full analysis: What’s Behind Gemini 4 Carbon’s One-Point Lead In The AI Rankings? on ThorstenMeyerAI.com
Get smart everyday buys delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
Gemini 4 Carbon scored 59 on Artificial Analysis’ Intelligence Index v5, one point ahead of Claude Opus 5.5 and two ahead of GPT-6 Astra. It leads four of the nine tests but trails rivals on five, and Google has not made it publicly available. The ranking measures performance on a specific set of tasks, not a universal measure of AI capability.
Google’s Gemini 4 Carbon has taken the top spot on Artificial Analysis’ Intelligence Index v5, scoring 59 points—one ahead of Claude Opus 5.5 and two ahead of GPT-6 Astra. The result puts Carbon first on this particular nine-test benchmark, but its lead is narrow, its performance varies by task, and Google is initially offering the model only to a limited group of cyber defenders.
The index assigns Carbon 59 points, compared with 58 for Claude Opus 5.5, 57 for GPT-6 Astra, 55 for GPT-6.1 Sol and 50 for Google’s previous model, Gemini 4 Argon. Artificial Analysis combines results from nine tests, divided into four equally weighted areas: knowledge work, coding, science and general performance. The ranking reflects models at particular settings on those tests; it is not a single, all-purpose measure of intelligence.
Carbon places first on four of the nine tests, according to the supplied results: hard reasoning and knowledge questions, science workflows conducted in a terminal, a test that penalizes incorrect answers, and business workflows using software tools. Its reported scores include 69% on HLE-Diamond, against 68% for Claude Opus 5.5, and 65% on Terminal-Bench-Science, against 63% for GPT-6 Astra. On AutomationBench-AA, Carbon is fractionally ahead of Gemini 4 Argon.
It trails competitors on the other five tests. Claude models lead office-style knowledge work, Claude Opus 5.5 leads real-world coding, GPT-6 Astra leads a physics research test, Kimi K3.1 Pro leads terminal tasks, and Meta’s Muse Wonder Full leads document reasoning. Carbon scored 1,609 Elo on AA-Briefcase, below Claude Sonnet 5.5 at 1,812 and Claude Opus 5.5 at 1,800. On FrontierCode 2.0, it scored 41%, compared with 46% for Claude Opus 5.5.
Cost and access also qualify the headline. Artificial Analysis puts the cost of running the index tasks with Carbon at $7.76 per task at standard prices. That is below Claude Opus 5.5’s $10.04, but above GPT-6 Astra’s $6.24 and well above GPT-6.1 Sol’s $1.27. Google’s initial API pricing is listed as $2 per million input tokens and $10 per million output tokens, with later prices of $4 and $20, respectively. The source gives no date for the broader release.
Infographic 01 · What the #1 is made of
Number one overall. First in four of nine tests.
The Intelligence Index is a decathlon score: nine tests, one number. Carbon wins on reasoning, science and careful knowledge, and trails on office work and coding.
The nine tests behind the score
Carbon at its highest setting (high) against the best model on each test.
| Test | Weight | Carbon | Leader or runner-up |
|---|---|---|---|
| HLE-Diamond Hard reasoning + knowledge questions | Weight 10% | Carbon 69% | Carbon leads Next: Claude Opus 5.5, 68% |
| Terminal-Bench-Science Science research work in a terminal | Weight 15% | Carbon 65% | Carbon leads Next: GPT-6 Astra, 63% |
| AA-Omniscience v1.1 Knowledge, wrong answers penalized | Weight 15% | Carbon 54 | Carbon leads Next: Claude Opus 5.5, 48 |
| AutomationBench-AA Business workflows across apps | Weight 5% | Carbon 78% | Carbon leads Next: Gemini 4 Argon, 78% |
| AA-Briefcase Office-style knowledge work | Weight 15% | Carbon 1,609 Elo | Behind Claude Sonnet 5.5, 1,812 |
| FrontierCode 2.0 Code a maintainer would merge | Weight 15% | Carbon 41% | Behind Claude Opus 5.5, 46% |
| CritPt v2 (hard) Research-level physics | Weight 10% | Carbon 63% | Behind GPT-6 Astra, 71% |
| Terminal-Bench 4.0 Agentic terminal tasks | Weight 10% | Carbon 65% | Behind Kimi K3.1 Pro, 67% |
| GDP.pdf Reasoning over documents | Weight 5% | Carbon 28% | Behind Muse Wonder Full, 33% |
The surprise in the knowledge test
- Accuracy: Claude Fable 5.1 68%, Claude Opus 5.5 66%, GPT-6.1 Sol and GPT-6 Astra 62%, Carbon 58%
- Hallucination rate: Carbon 11%, Claude Opus 5.5 and GPT-6 Astra about 50%
- Carbon wins because it rarely bluffs, not because it knows the most
Read the weights
- Each of the four blocks counts 25%: knowledge work, coding, science, general
- Carbon trails in two of the four heaviest tests (15% each): office-style work and real-world code
- Hallucination rate = wrong answers among all questions not answered correctly
Source: Artificial Analysis, 28 Oct 2026; Index v5 announcement, 27 Oct 2026; Artificial Analysis model data, retrieved 29 Oct 2026. Scores rounded as in the article; Carbon tested at high, other models at their highest tested setting.
Powered by Thorsten Meyer AI
A One-Point Lead Has Limits
Carbon’s result matters because it shows how a model can top a combined benchmark without leading across most of its individual tests. The one-point margin over Claude Opus 5.5 is small, while the models’ relative strengths differ across coding, research, office work and other tasks. For businesses choosing a system, the overall rank may be less useful than results on the work they actually need done.
The cost figures add another practical constraint. A model that performs well on a demanding benchmark may still be a poor fit if it takes more time or money to complete a company’s particular task. The supplied source advises organizations to test models against their own work and compare cost per completed task, rather than selecting on rank alone. These benchmark costs are reported estimates, not a guarantee of what a deployment will cost.
For consumers, the immediate effect is limited: Carbon is not publicly available under the rollout described. The reported advantage is most relevant to difficult, extended tasks, rather than a claim that everyday chat will improve for users now.
Products Worth Considering
HP Sprocket Photo Booth Machine Instant Color Photo Printer with 10.1 Touchscreen, LED Ring Light, 250+ Frames, 3×4 ZINK Paper, QR Code Shar
FLEXISPOT Standing Desk with Monitor Riser & Cable Management
Apple iPad Air 11-inch (M2), 512GB, Wi-Fi, Space Gray
As an affiliate, we earn on qualifying purchases.
How Index v5 Builds Its Score
Artificial Analysis introduced version 5 on October 27, 2026, according to the source. The index gives equal weight to four categories: knowledge work, coding, science and general performance. Within those categories, individual tests carry different weights; for example, AA-Briefcase and FrontierCode 2.0 each account for 15% of the overall score.
The source says 60% of the index uses private test sets or held-out answers, intended to reduce the chance that models have memorized test questions. Many tasks are also designed to involve longer workflows, such as operating in a terminal, producing reports or fixing code. That makes the benchmark different from a simple question-and-answer exam, but it still represents a selected set of tasks and scoring choices.
One result illustrates how those choices affect rankings. On AA-Omniscience, Carbon scores 54 under a system that penalizes wrong answers. Its reported plain accuracy is 58%, below Claude Fable 5.1 at 68%, Claude Opus 5.5 at 66%, and GPT-6.1 Sol and GPT-6 Astra at 62% each. Artificial Analysis reports an 11% hallucination rate for Carbon on this test, compared with about 50% for Claude Opus 5.5 and GPT-6 Astra. The source describes the measure as the share of wrong answers among questions the model did not get right. Thus, Carbon’s advantage on this test is attributed to fewer incorrect guesses, not the highest number of correct answers.
““Number one” on one index is a narrower claim than it sounds.”
— Thorsten Meyer, ThorstenMeyerAI.com
What the Ranking Cannot Settle
The supplied material does not establish how Carbon will perform across the full range of real-world uses, or whether the one-point advantage will persist as tests and model versions change. Its score depends on the nine selected benchmarks, their relative weights and the settings used to run each model. Different tasks or scoring choices could produce a different order.
Public access is also unresolved. Google is described as making Carbon available first to a subset of cyber defenders, then to trusted partners, developers, businesses and consumers, but no release dates are given for those stages. The supplied source ends mid-sentence in its discussion of a separate coding-agent comparison, so that claim cannot be completed or verified from the material provided.
The reported test costs and API prices do not show what a company will spend on its own workload. Actual use can depend on task length, input and output volumes, model settings and other deployment choices. The source also does not provide enough detail here to independently assess how the index’s cost-per-task figure was calculated.
Google’s Staged Rollout
Google’s reported plan is to expand access from a subset of cyber defenders to trusted partners, then developers, businesses and consumers. The company has not supplied dates for those stages in the source material. Until public access begins, most users cannot test Carbon directly or judge whether its benchmark strengths translate to their needs.
For organizations evaluating the model, the next useful evidence will be results on their own workflows, including accuracy, completion time, reliability and total cost. Further comparisons may also clarify whether Carbon’s lead holds across other evaluations and how it performs when paired with tools. For now, the confirmed development is a narrow first-place ranking—not a broad verdict on which AI model is best for every user.
Key Questions
What is Gemini 4 Carbon’s score on the index?
It scored 59 points on Artificial Analysis’ Intelligence Index v5, one point ahead of Claude Opus 5.5 at 58.
Does Carbon lead all nine tests?
No. The supplied results say Carbon places first on four of nine tests and trails other models on the remaining five.
Can the public use Gemini 4 Carbon now?
Not according to the source. Google is reported to be starting with a subset of cyber defenders, followed by other groups, but no public-release date is specified.
Does first place mean Carbon is the best model for every task?
No. The index combines a particular set of nine tests. The reported results show other models ahead in areas including office-style knowledge work, coding, physics research, terminal tasks and document reasoning.
How much does it cost to run the benchmark tasks with Carbon?
Artificial Analysis’ reported estimate is $7.76 per index task at standard prices. That is a benchmark estimate, not a prediction of an organization’s costs on its own workload.
Source: ThorstenMeyerAI.com
