🔍 Read the full analysis: A New #1 AI Model? What Gemini 4 Carbon’s One-Point Score Really Shows on ThorstenMeyerAI.com
Get the little things that make your day 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 leads the Artificial Analysis Intelligence Index v5 with 59 points, one ahead of Claude Opus 5.5 and two ahead of GPT-6 Astra. It ranks first on four of nine tests, but trails competitors on five and is not yet publicly available. The result measures performance on a particular set of tasks, not a universal ranking of AI capability.
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. The narrow lead applies to nine benchmark tests at particular model settings; Carbon is not yet publicly available, and the result does not establish that it is best for every task.
Artificial Analysis, an independent benchmarking company, launched version 5 of its index on October 27. It combines results from nine tests into a single score. The reported rankings are 59 for Gemini 4 Carbon, 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. Those figures compare models under the test conditions and settings used for the index.
Carbon places first on four of the nine tests: hard reasoning and knowledge questions, science workflows conducted in a terminal, knowledge questions scored with a penalty for wrong answers, and business workflows across software tools. Its results 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, it scores 78%, slightly ahead of its predecessor, according to the source report.
The model places behind competitors on the other five tests. Claude models lead on office-style knowledge work, Claude Opus 5.5 leads on real-world coding, GPT-6 Astra leads on physics research, Kimi K3.1 Pro on terminal tasks and Meta’s Muse Wonder Full on document reasoning. The report puts Carbon’s score on the office-style AA-Briefcase test at 1,609 Elo, below Claude Sonnet 5.5 at 1,812 and Claude Opus 5.5 at 1,800. On FrontierCode 2.0, Carbon scores 41%, compared with 46% for Claude Opus 5.5.
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
The index gives organizations a comparable snapshot of performance across several kinds of work, but its one-point gap at the top is not a guarantee that Carbon will outperform rivals in a company’s actual workflow. Carbon leads on some science, reasoning and business-process tests while losing on other work that many teams may rely on, including office tasks and coding.
Price and time also matter. The source report estimates $7.76 per index task for Carbon at standard prices. That is about 23% less than Claude Opus 5.5 at $10.04, but about 1.2 times GPT-6 Astra at $6.24 and roughly six times GPT-6.1 Sol at $1.27. These are reported benchmark-task costs, not a universal estimate for every customer’s usage. Organizations would need to test their own workloads and compare cost per completed task before making a switch.
For everyday users, the immediate effect is limited: Carbon is not publicly accessible. If it becomes broadly available, the reported strengths may matter most on demanding, multi-step work rather than routine chat. The index cannot by itself establish how those tasks will perform in a consumer product or a particular business environment.
Products Worth Considering
Hyperice Normatec Elite Wireless Compression Boots
Bowflex SelectTech 840 Adjustable Kettlebell
Garmin Venu 3 Fitness Smartwatch, 45 mm, Black
As an affiliate, we earn on qualifying purchases.
What the New Index Tests
Artificial Analysis describes the Intelligence Index as a combined score from nine tests arranged in four equally weighted groups: knowledge work, coding, science and general ability. The reported weights are 25% for each group. Individual tests have different weights within them, so a model’s overall position reflects performance across the index’s chosen mix, not just its strongest result.
Version 5 uses private test sets or held-out answers for 60% of its evaluations, according to the source report. Many tasks are extended agent-style activities, such as working in a terminal, producing reports or fixing code. That design is intended to test more than short-form answers, though it also makes runs more expensive. The index remains a benchmark: changing the tasks, their weights or the model settings could change the ranking.
Carbon’s result on AA-Omniscience illustrates why a single score needs interpretation. It earns 54 points on that test, six ahead of Claude Opus 5.5, despite answering fewer questions correctly in the plain-accuracy figures reported by the source. Carbon’s accuracy is 58%, compared with 66% for Claude Opus 5.5 and 62% for GPT-6 Astra. The test also penalizes wrong answers, and the report gives Carbon an 11% hallucination rate, versus about 50% for those two competitors. In this scoring setup, abstaining rather than guessing can improve a model’s result.
Access, Costs and Comparisons
No public release date is given for Gemini 4 Carbon. The reported rollout is staged: a subset of cyber defenders first, then trusted partners, followed by developers, businesses and consumers. The source does not specify how large the initial group is, what eligibility rules apply, or when each later stage will begin.
The report lists introductory API pricing of $2 per million input tokens and $10 per million output tokens, later rising to $4 and $20 respectively. It does not give a date for the price change or enough detail to establish how those rates translate to cost for a customer’s particular workload. The index-task estimate and token prices describe different measures and should not be treated as interchangeable.
The supplied report also begins describing a separate coding-agent comparison, but its account cuts off mid-sentence. It gives a score of 62 for Claude Code with Claude Opus 5.5 and 59 for Google’s Antigravity CLI with Carbon, while the explanation of Artificial Analysis’s assessment is incomplete. That comparison should not be read as a complete account of the company’s findings from the available material.
Google’s Staged Rollout
The next development is whether Google begins the planned rollout beyond the initial subset of cyber defenders. The company’s stated sequence is trusted partners, then developers and businesses, and later consumers, but no timetable is available in the source. Public availability, product integrations and any changes to API pricing remain open questions.
For companies considering Carbon, the practical next step is to compare it with alternatives on representative work: accuracy, willingness to abstain, completion time and the full cost of a successful task. For readers, the index result is a benchmark update rather than an immediate product change. Further rankings or broader access could show whether Carbon’s narrow lead holds across new tests and real-world use.
Key Questions
Is Gemini 4 Carbon the best AI model?
It ranks first on the Artificial Analysis Intelligence Index v5, with 59 points. That is a result on this index’s nine tests, not proof that Carbon is best for every task or use case.
How far ahead is Carbon?
Carbon scores one point ahead of Claude Opus 5.5 and two ahead of GPT-6 Astra in the reported ranking. The source does not provide evidence that this small difference alone predicts performance on a specific customer’s work.
Can the public use Gemini 4 Carbon now?
No public access is reported yet. Google is starting with a subset of cyber defenders, then plans to expand to trusted partners, developers, businesses and consumers; the source gives no dates for those stages.
Why does Carbon lead the knowledge test if it answers fewer questions correctly?
AA-Omniscience penalizes wrong answers as well as measuring knowledge. The report says Carbon has lower plain accuracy than several rivals but a much lower reported hallucination rate, so avoiding guesses helps its score on that test.
How much does Carbon cost?
The source reports introductory API prices of $2 per million input tokens and $10 per million output tokens, later rising to $4 and $20. It also estimates $7.76 per index task at standard prices; actual costs depend on usage and the pricing terms available to customers.
Source: ThorstenMeyerAI.com
