🔍 Read the full analysis: Can Gemini 4 Carbon’s One-Point AI Lead Settle The Model Debate? on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Artificial Analysis’s Intelligence Index v5 puts Google’s Gemini 4 Carbon at 59 points, one ahead of Claude Opus 5.5 and two ahead of GPT-6 Astra. The ranking is based on nine tests, and Carbon leads four while falling behind on five; it is not yet publicly available.
Google’s Gemini 4 Carbon has taken the top position on the Artificial Analysis Intelligence Index v5, scoring 59 points—one more than Claude Opus 5.5 and two more than GPT-6 Astra. The result is a narrow lead on one benchmark, not a general finding that Carbon is the best model for every task; Google says access is currently limited to a subset of cyber defenders.
The index combines results from nine tests grouped into knowledge work, coding, science and general capabilities, with each group accounting for a quarter of the total. Artificial Analysis launched version 5 on October 27. The source report says 60% of the index uses private tests or held-out answers, intended to reduce the likelihood that models have memorized evaluation material.
Carbon places first on four of the nine tests: HLE-Diamond, Terminal-Bench-Science, AA-Omniscience and AutomationBench-AA. Its reported scores include 69% on HLE-Diamond, one point above Claude Opus 5.5, and 65% on Terminal-Bench-Science, two points ahead of GPT-6 Astra. On AutomationBench-AA, it edges its predecessor, Gemini 4 Argon.
It ranks below competitors on the other five tests. Claude models lead on office-style knowledge work, Claude Opus 5.5 is ahead on real-world coding, GPT-6 Astra leads in physics research, and Meta’s Muse Wonder Full leads on document reasoning. The report puts Carbon’s cost at $7.76 per index task at standard prices: less than Claude Opus 5.5 at $10.04, but more than GPT-6 Astra at $6.24 and far above GPT-6.1 Sol at $1.27.
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, Not a Universal Win
The one-point margin is meaningful as a result on this particular index, but it is small enough that readers should focus on the underlying task scores as well as the overall ranking. Since the index averages different kinds of work, a model’s position can reflect the benchmark’s chosen tasks and weights. A business evaluating coding, document analysis or office work may care more about the relevant test—and its own results—than the overall score.
The cost figures also change the practical comparison. Carbon is cheaper per index task than Claude Opus 5.5 under the listed standard prices, but pricier than GPT-6 Astra and substantially more expensive than GPT-6.1 Sol. Those figures are benchmark-task costs, not a guarantee of what any particular customer will pay for a workload. The report advises companies to test models on their own tasks and compare the cost per completed job rather than choosing by rank alone.
For individual users, there is no immediate change: Carbon is not publicly available. If access expands, its benchmark profile suggests the reported advantage may be more relevant to difficult, extended tasks than to routine chat. That is an interpretation of the test results, not a guarantee of how it will perform for each user.
Products Worth Considering
Amazon Kindle Paperwhite (12th Generation, 2024), 16 GB, 7-inch Display, Jade
Samsung Galaxy Tab S10 FE 128GB Wi-Fi Android Tablet, Blue
Anker SOLIX S2000 Portable Power Station, 2,010Wh, 1,500W
As an affiliate, we earn on qualifying purchases.
How the Index Produces Its Score
Artificial Analysis describes its index as a combined score across nine evaluations, not a single test of general intelligence. In version 5, the four equal-weight blocks are knowledge work, coding, science and general performance. Individual tests within those blocks carry different weights, so a strong result in one area does not determine the overall ranking on its own.
The report highlights AA-Omniscience as an example of how a test’s scoring rules affect rankings. Carbon scores 58% on plain accuracy, below Claude Fable 5.1 at 68%, Claude Opus 5.5 at 66%, and GPT-6.1 Sol and GPT-6 Astra at 62%. But the evaluation penalizes incorrect answers, and Carbon’s reported hallucination rate—defined in the source as wrong answers among questions the model did not get right—is 11%, compared with about 50% for Claude Opus 5.5 and GPT-6 Astra. Carbon therefore leads that test despite not answering the most questions correctly.
Carbon’s score of 59 also compares with 50 for Gemini 4 Argon, Google’s previous model in the ranking. The source report says Carbon improves on Argon in office-style knowledge work but still trails leading Claude models there. The overall index is one view of model performance; its results do not establish how every model will behave across all products, settings or real-world tasks.
“A subset of cyber defenders”
— Google, as reported by ThorstenMeyerAI.com
Access and Real-World Performance
Google has not given a public release date for Carbon. The reported rollout plan is to start with a subset of cyber defenders, then extend to trusted partners, developers, businesses and consumers, but the timing and eligibility for each stage are not specified.
The source material also does not establish how Carbon will compare with rivals on each customer’s workload, or whether the benchmark cost figures will match actual usage costs. Index positions depend on the selected tests, their weights and the model settings evaluated. The one-point gap should not be read as proof of a consistent advantage across products or tasks.
The report provides introductory API prices of $2 per million input tokens and $10 per million output tokens, later rising to $4 and $20, respectively. It does not specify when those prices take effect or provide further terms for the rollout.
Google’s Staged Rollout Ahead
The next reported step is expanded access from the initial cyber-defender group to trusted partners, followed by developers, businesses and consumers. No schedule has been announced, so it remains unclear when general users will be able to try Carbon.
For companies weighing adoption, the immediate next step is to run relevant tasks against their own work and compare accuracy, completion rates, response time and cost. Artificial Analysis’s overall score provides a benchmark comparison, but Carbon’s mixed results across coding, office work, science and knowledge tests make task-specific evaluation especially relevant. Further public access and independent testing may clarify how its benchmark profile translates into ordinary use.
Key Questions
What is Gemini 4 Carbon’s score on the index?
It scores 59 points on Artificial Analysis Intelligence Index v5, one point ahead of Claude Opus 5.5 and two ahead of GPT-6 Astra.
Does the ranking show Carbon is the best AI model overall?
No. It shows Carbon leading this particular index, which combines nine tests. Carbon ranks first on four tests and behind other models on five, so the result does not establish that it is best for every task.
Why does Carbon lead the Omniscience test if other models answer more correctly?
The test penalizes wrong answers. The report says Carbon has lower plain accuracy than several rivals but a much lower reported hallucination rate, which helps it score higher under that test’s scoring rules.
Can the public use Gemini 4 Carbon now?
No public release is reported. Google is initially providing access to a subset of cyber defenders and plans to expand to other groups, but it has not announced dates.
How much does Carbon cost to run?
The report estimates $7.76 per index task at standard prices. It also lists introductory API rates of $2 per million input tokens and $10 per million output tokens, later rising to $4 and $20; the timing of that change is not stated.
Source: ThorstenMeyerAI.com
