🔍 Read the full analysis: How Much Does Gemini 4 Carbon’s One-Point AI Lead Really Mean? 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’s Intelligence Index v5, one point ahead of Claude Opus 5.5 and two ahead of GPT-6 Astra. The result reflects performance across nine selected tests, not a universal measure of AI capability; Carbon leads four tests and trails on five. Google has not made the model publicly available, and its reported API prices and limited rollout remain relevant to any comparison.
Google’s Gemini 4 Carbon ranks first on Artificial Analysis’s newly rebuilt Intelligence Index v5, scoring 59 points—one ahead of Claude Opus 5.5 and two ahead of GPT-6 Astra. The result is a narrow lead across a specific set of nine tests, not a finding that Carbon is best at every AI task; Google is also limiting access to a small group of cyber defenders for now.
Artificial Analysis launched version 5 of its index on October 27, according to the source report, and Carbon took the top position the following day. Claude Opus 5.5 scored 58, GPT-6 Astra 57, GPT-6.1 Sol 55 and Google’s earlier Gemini 4 Argon 50. The index combines nine tests, grouped into four equally weighted areas: knowledge work, coding, science and general capability.
Carbon placed first on four of the nine tests: HLE-Diamond, Terminal-Bench-Science, AA-Omniscience and AutomationBench-AA. It was behind other models on the remaining five, including office-style knowledge work, real-world coding, physics research, terminal tasks and document reasoning. The index’s overall score compresses those differing results into one ranking, so the one-point gap does not mean Carbon won every category or that its lead is equally strong across tasks.
The report estimates a cost of $7.76 per index task at standard prices. That is below the stated $10.04 for Claude Opus 5.5 at its maximum setting, but above GPT-6 Astra’s $6.24 and substantially above GPT-6.1 Sol’s $1.27. It also reports introductory API pricing of $2 per million input tokens and $10 per million output tokens, rising later to $4 and $20 respectively. The report does not specify a date for that later pricing or give a public launch date.
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 ranking may help businesses identify models worth testing, but it is not a direct answer to which system will perform best in their own work. Task mix and cost per completed task matter: Carbon’s four test wins could be relevant for some research, business-tool or cautious question-answering workflows, while its weaker office-work and coding results may matter more to other buyers.
The index’s scoring also rewards a particular kind of reliability. In AA-Omniscience, Carbon scores 54, ahead of Claude Opus 5.5 by six points, even though the source says several Claude and GPT models answer more questions correctly. Artificial Analysis’s measure penalizes wrong answers, and the report gives Carbon an 11% hallucination rate, compared with about 50% for Claude Opus 5.5 and GPT-6 Astra. That distinction could matter where unsupported answers are costly, but it does not establish that Carbon is more accurate on every task.
For everyday users, the immediate effect is limited: Carbon is not publicly available according to the report. For companies, the headline is a reason to evaluate the model against their own workloads—not, by itself, a reason to switch providers.
Products Worth Considering
LG UltraGear 27-Inch QHD OLED Gaming Monitor, 240Hz, 0.03ms, G-SYNC Compatible, FreeSync Premium Pro, HDMI 2.1
FLEXISPOT EN1-PRO Electric Standing Desk, 79 x 32, Black Walnut
Samsung 27 Odyssey OLED G5 (G50SF) QHD QD-OLED Gaming Monitor
As an affiliate, we earn on qualifying purchases.
What the New Index Tests
Artificial Analysis describes its Intelligence Index as a combined score from nine benchmark tests. In v5, the tests sit within four equally weighted blocks. Knowledge work includes AA-Briefcase, AutomationBench-AA and GDP.pdf; coding includes FrontierCode 2.0 and Terminal-Bench 4.0; science includes Terminal-Bench-Science and the hard subset of CritPt v2; and general capability includes AA-Omniscience and HLE-Diamond.
The report says 60% of the index uses private test sets or held-out answers, and many tasks require models to work over longer sequences, such as completing terminal-based jobs, producing reports or fixing code. Those design choices aim to reduce the influence of memorized questions and test more involved work. They do not make the index a complete measure of general intelligence: its result depends on the tests selected, their weights and the model settings used.
Carbon’s individual results illustrate why the aggregate ranking needs qualification. On HLE-Diamond, it scores 69% against Claude Opus 5.5’s 68%; on Terminal-Bench-Science, it records 65% against GPT-6 Astra’s 63%. But on AA-Briefcase, the report lists Carbon 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.
“Choose by task and by cost per finished task, not by rank.”
— Thorsten Meyer, source report author
Access, Costs and Test Scope
The source report says Google plans to offer Carbon first to a subset of cyber defenders, then to trusted partners, developers, businesses and consumers. It gives no timetable for those stages, and the model is not yet publicly available. The reported token prices are also described as introductory, with later rates listed but no effective date supplied.
The report provides index scores and selected task results, but readers cannot infer from the overall score alone how Carbon will perform on a particular organization’s data, tools or safety requirements. It also does not include enough detail here to independently assess every benchmark’s scoring procedure or reproduce the results. The ranking is therefore confirmed as a reported result of this index, while its relevance to other tasks remains uncertain.
The Next Test Is Availability
The next practical milestone is wider access to Gemini 4 Carbon. Google’s stated rollout sequence begins with cyber defenders and proceeds through trusted partners to developers, businesses and consumers, but no dates are given. Until access expands, most prospective users cannot test the model directly or compare its performance with alternatives on their own workloads.
When businesses can evaluate it, useful comparisons will include task success, error rates, speed and the full cost of completing work—not only per-token rates or the index position. Further benchmark details, updated pricing and a public release schedule could change how buyers assess the result. For now, the confirmed development is a narrow first-place score on one index, paired with restricted access and mixed results across its component tests.
Key Questions
How far ahead is Gemini 4 Carbon?
Carbon scored 59 points on Artificial Analysis’s Intelligence Index v5, one point ahead of Claude Opus 5.5 at 58 and two ahead of GPT-6 Astra at 57.
Does the result mean Carbon is the best model at every task?
No. The source report says Carbon leads four of nine tests and trails on five, including office-style knowledge work and real-world coding. The index score summarizes performance on its selected tests.
Why does Carbon lead AA-Omniscience if other models answer more correctly?
According to the report, AA-Omniscience penalizes wrong answers. Carbon scores lower on plain accuracy than several listed models but has a reported 11% hallucination rate, compared with about 50% for Claude Opus 5.5 and GPT-6 Astra. The test therefore rewards its lower rate of wrong guesses.
Can the public use Gemini 4 Carbon now?
Not according to the October 29 report. Google is initially providing access to a subset of cyber defenders, with wider groups listed for later stages; no public release date is specified.
What does the reported cost comparison show?
The report estimates $7.76 per index task at standard prices, compared with $10.04 for Claude Opus 5.5, $6.24 for GPT-6 Astra and $1.27 for GPT-6.1 Sol. These are reported benchmark-task costs, not a guarantee of what a particular user’s work will cost.
Source: ThorstenMeyerAI.com
