🔍 Read the full analysis: Is Gemini 4 Carbon’s #1 AI Ranking More Close Call Than Clear Verdict? on ThorstenMeyerAI.com
Get the latest gadgets 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 placed first on four of nine tests, while its performance varies by task, its test-run cost is higher than some rivals, and Google has not made it generally available.
Google’s Gemini 4 Carbon has taken the top position 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 is a narrow lead on one benchmark, not an overall verdict across every kind of AI work: Carbon ranked first on four of nine tests, and Google has not yet released it for general use.
Artificial Analysis’ index combines results from nine tests into a single score. In the figures reported by ThorstenMeyerAI.com, Gemini 4 Carbon in its high-reasoning configuration scored 59; Claude Opus 5.5 at maximum scored 58; GPT-6 Astra at maximum scored 57; and GPT-6.1 Sol at maximum scored 55. Google’s previous model, Gemini 4 Argon, scored 50. These are index results for the tested configurations, not a claim that Carbon leads on every task.
Carbon finished first on HLE-Diamond, Terminal-Bench-Science, AA-Omniscience and AutomationBench-AA. It lost the reported comparisons in office-style knowledge work, real-world coding, physics research, general terminal tasks and document reasoning. For example, Carbon scored 41% on FrontierCode 2.0, against 46% for Claude Opus 5.5; in AA-Briefcase, its 1,609 Elo was below Claude Sonnet 5.5’s 1,812 and Claude Opus 5.5’s 1,800.
The source reports an estimated $7.76 per index task at standard prices. That is below Claude Opus 5.5’s $10.04, but above GPT-6 Astra’s $6.24 and far above GPT-6.1 Sol’s $1.27. Google’s introductory API price is listed as $2 per million input tokens and $10 per million output tokens, later rising to $4 and $20 respectively. These pricing figures describe different measures: the API rates are token prices, while the index-task figure estimates benchmark usage.
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 gives buyers a useful signal that Carbon is competitive across a broad set of demanding evaluations. But a one-point margin over Claude Opus 5.5 is small, and the results show meaningful differences between tasks. A company whose work is mostly coding or office-document production may get a different answer from one focused on science workflows or business automation.
The cost figures also complicate a simple “number one” choice. A model that performs well on a benchmark may still be a poor fit if it costs more to complete a company’s actual work, takes too long, or makes errors that require human review. The source describes Carbon as slow and advises companies to test models on their own tasks and compare the cost per completed job, rather than selecting solely by index position.
For everyday users, there is no immediate product change because Carbon is not publicly available. If it is later released broadly, the source expects its advantage to be most relevant on long, difficult tasks rather than routine chat. That is an interpretation, not a published guarantee about how consumers will experience the model.
Popular right now
Bose QuietComfort Wireless Bluetooth Headphones with Active Noise Cancellation, Mic, USB‑C Charging, Deep Bass, 24‑Hour Battery – White Smok
iRobot Roomba Max 715 Pet Robot Vacuum with AutoEmpty Dock
Samsung 42-Inch S90H OLED 4K Smart TV Bundle
As an affiliate, we earn on qualifying purchases.
How Artificial Analysis Builds Its Index
Artificial Analysis launched version 5 of its Intelligence Index on October 27, 2026, according to the source. It groups nine evaluations into four equally weighted categories: knowledge work, coding, science and general performance. The individual tests do not all carry the same weight. For example, AA-Briefcase and FrontierCode 2.0 each account for 15% of the overall score, while several other tests count for 5% or 10%.
The evaluations include agent-style work such as using terminal tools, preparing reports and handling code, alongside tests of knowledge and reasoning. The source says 60% of v5 uses private test sets or held-out answers, intended to reduce the effect of models having memorized test questions. The index nevertheless measures a defined set of tasks and model settings; changing the tests or their weights could change the order.
One of Carbon’s reported wins illustrates how scoring rules shape rankings. On AA-Omniscience, Carbon scored 54 points, ahead of Claude Opus 5.5 by six. Yet Carbon’s reported plain accuracy was 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%. The test penalizes wrong answers, and the source reports an 11% hallucination rate for Carbon versus about 50% for Claude Opus 5.5 and GPT-6 Astra. Thus, Carbon’s higher score reflects its lower tendency to give incorrect answers as well as what it knows.
““Number one” on one index is a narrower claim than it sounds.”
— Thorsten Meyer, ThorstenMeyerAI.com
Availability and Benchmark Caveats
No date for general release is provided. Google’s reported rollout order begins with a subset of cyber defenders, followed by trusted partners, developers, businesses and consumers, but the timing and eligibility at each stage remain unspecified.
The supplied report does not include the complete underlying Artificial Analysis results, testing methodology or full list of rival scores for every task. Its discussion of the separate Coding Agent Index is also cut off mid-sentence. The reported scores and costs should therefore be treated as figures attributed to the source, not independently verified here. The index’s single composite score also cannot establish which model will perform best on a particular organization’s work.
Google’s Staged Rollout Ahead
According to the source, Google plans to move Carbon from a limited initial group of cyber defenders to trusted partners, then to developers, businesses and consumers. No schedule was supplied, so the timing of those stages—and when ordinary users might access the model—remains unknown.
Artificial Analysis’ reported v5 ranking is the current comparison in this account. For companies considering a switch, the practical next step is to test Carbon against their own tasks, including quality, completion time and total cost. Until wider access and those evaluations are available, the one-point index lead is evidence of a close benchmark result, not a settled choice for every user.
Key Questions
What score did Gemini 4 Carbon receive?
It scored 59 points on Artificial Analysis’ Intelligence Index v5, according to the supplied report. Claude Opus 5.5 scored 58 and GPT-6 Astra scored 57.
Did Carbon win all nine tests?
No. The report says Carbon placed first on four of nine tests. It trailed rivals in areas including office-style knowledge work and real-world coding.
Is Gemini 4 Carbon available to the public?
Not yet, according to the source. Google is reportedly starting with a subset of cyber defenders and plans to expand access in stages, but it has not provided a public release date.
Does Carbon know more than its rivals?
Not on every reported measure. The source says several Claude and GPT models had higher plain accuracy on AA-Omniscience, while Carbon scored better on that test because it was less likely to give a wrong answer.
How should businesses interpret the ranking?
The index is one comparison across a defined set of evaluations, not a guarantee of performance on a company’s work. The report recommends testing models on real tasks and comparing the cost per completed task.
Source: ThorstenMeyerAI.com
