🔍 Read the full analysis: Gemini 4 Carbon’s New #1 AI Model Title Comes Down To One Point on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
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 index’s nine tests, but loses five, carries a stated standard index-task cost of $7.76, and is not yet publicly available.
Google’s Gemini 4 Carbon ranks first on the Artificial Analysis Intelligence Index v5 with 59 points, one point ahead of Claude Opus 5.5 and two ahead of GPT-6 Astra, according to the benchmark results reported October 29. The title is limited to this index: Carbon leads four of its nine tests, loses the other five and is not yet publicly available.
The v5 index combines nine evaluations into a single score, with four categories—knowledge work, coding, science and general ability—each assigned 25% of the total weight. In the reported ranking, Claude Opus 5.5 scored 58, GPT-6 Astra 57, GPT-6.1 Sol 55 and Google’s previous Gemini 4 Argon 50. The scores compare particular model versions and settings on this benchmark; they are not a universal measure of AI capability.
Carbon places first on four tests: HLE-Diamond, Terminal-Bench-Science, AA-Omniscience and AutomationBench-AA. Its reported results include 69% on HLE-Diamond, compared with 68% for Claude Opus 5.5, and 65% on Terminal-Bench-Science, compared with 63% for GPT-6 Astra. On AutomationBench-AA, it scores 78%, slightly ahead of Gemini 4 Argon.
Its weaker results are also part of the picture. Carbon trails Claude models on office-style knowledge work and real-world coding, GPT-6 Astra on a physics research test, Kimi K3.1 Pro on terminal tasks and Meta’s Muse Wonder Full on document reasoning. The source reports a standard cost of $7.76 per index task: below Claude Opus 5.5 at $10.04, above GPT-6 Astra at $6.24 and well above GPT-6.1 Sol at $1.27. These figures describe the benchmark’s stated task costs, not a guaranteed price for every workload.
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 Narrow Lead With Practical Trade-Offs
The one-point margin makes the ranking sensitive to what is measured and how the tests are weighted. A company choosing a model for office documents, coding or research could get a different result from the overall leaderboard because Carbon does not lead across all those areas. The index offers a comparison across a defined set of tasks, not a simple answer to which model is best for every user.
Cost and access also affect the immediate value of the result. Artificial Analysis’ tasks can involve extended agent work, so a per-task benchmark price may help organizations think beyond a model’s headline score. But the supplied figures do not establish what a particular company will pay on its own workload. Google’s staged rollout means that, for now, most developers, businesses and consumers cannot test Carbon directly.
For readers using consumer AI products, the ranking does not signal a change they can make today. If the rollout reaches broader users, the reported strengths suggest the difference may matter more on difficult, multi-step tasks than on ordinary chat. That is an interpretation of the test results, not a guarantee of performance on individual tasks.
Popular right now
Sony BRAVIA 5 85-Inch Mini LED 4K HDR Smart TV (K-85XR50)
Samsung 85-Inch QN70H Neo QLED 4K Smart TV (2026)
De'Longhi 15-Bar Pump Espresso Machine with Milk Frother & Temperature Control
As an affiliate, we earn on qualifying purchases.
What the V5 Index Tests
Artificial Analysis launched the fifth version of its index on October 27, according to the source material. It groups its nine tests into four equally weighted blocks: knowledge work, coding, science and general ability. Individual tests have different shares within those blocks, so a high score on one evaluation does not by itself determine the overall result.
The source says 60% of v5 uses private test sets or held-out answers, intended to reduce the chance that models score well through memorizing public questions. Many evaluations involve longer tasks, such as working in a terminal, producing reports or changing code. Those features make the index relevant to agent-style work, but its results still depend on the selected tasks and scoring rules.
One result illustrates how those rules shape the ranking. On AA-Omniscience, Carbon scores 54, six points ahead of Claude Opus 5.5. Yet the source reports higher plain-answer accuracy for several rivals: Claude Fable 5.1 at 68%, Claude Opus 5.5 at 66%, and GPT-6.1 Sol and GPT-6 Astra at 62%, compared with Carbon at 58%. The test penalizes wrong answers, and the source reports an 11% hallucination rate for Carbon, against about 50% for Claude Opus 5.5 and GPT-6 Astra. In this evaluation, abstaining rather than guessing can improve a model’s score.
Limits of the One-Point Margin
The source does not provide full details of the index methodology, uncertainty ranges or whether a one-point difference is statistically meaningful. It also does not establish how the ranking would change under different test selections, category weights, model settings or pricing arrangements. The score confirms Carbon’s position on this version of this index, not a general consensus that it is the strongest model.
Access timing remains unsettled. Google’s reported sequence starts with a subset of cyber defenders, followed by trusted partners, then developers, businesses and consumers, but no dates are given. The supplied material also gives introductory API prices of $2 per million input tokens and $10 per million output tokens, later rising to $4 and $20, but does not state when those prices apply or detail the terms and availability for different users.
Some comparisons in the source are incomplete. It lists several tests Carbon loses and provides specific scores for some, but the available material ends during a discussion of coding tools. It therefore does not support a full accounting of every competitor’s result, or a complete assessment of the coding-agent comparison.
Rollout and Independent Testing
According to the source, Google plans to begin with a subset of cyber defenders, then extend access to trusted partners, developers, businesses and consumers. No launch schedule is specified, so the timing and conditions for each stage remain unknown.
For organizations considering the model, the most useful next step will be testing Carbon on representative tasks and comparing both performance and cost with available alternatives. The leaderboard can narrow the field, but it cannot establish how a model will perform on a company’s own documents, software or workflows. Further details on public availability, API terms and independent benchmark results would clarify how broadly the 59-point result translates beyond the index.
Key Questions
What does Gemini 4 Carbon’s first-place ranking mean?
It means Carbon scored 59 points on Artificial Analysis’ Intelligence Index v5, one point above Claude Opus 5.5. It does not mean Carbon leads every test or is best for every use.
Which tests does Carbon lead?
The source reports first place on HLE-Diamond, Terminal-Bench-Science, AA-Omniscience and AutomationBench-AA. Carbon loses the other five tests in the index.
Can the public use Gemini 4 Carbon now?
No public availability date is given. The reported rollout begins with a subset of cyber defenders, then moves to trusted partners and later to developers, businesses and consumers.
Is Carbon cheaper than its closest rivals?
The source gives a standard cost of $7.76 per index task. That is below Claude Opus 5.5’s $10.04, but above GPT-6 Astra’s $6.24 and GPT-6.1 Sol’s $1.27. Actual costs for other workloads are not established by those benchmark figures.
Why does Carbon win the omniscience test with lower accuracy?
AA-Omniscience penalizes wrong answers. The source says Carbon has lower plain-answer accuracy than several rivals but a lower reported hallucination rate, so it can score better by avoiding guesses when it does not know an answer.
Source: ThorstenMeyerAI.com
