🔍 Read the full analysis: A One-Point Gap Puts Gemini 4 Carbon At #1: What The AI Ranking Hides 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 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 in four of nine tests, but the result reflects a particular benchmark mix and model setting—not a universal measure of capability. Google says Carbon is initially available to a subset of cyber defenders; a broader release date has not been announced.
Google’s Gemini 4 Carbon has moved to the top of the Artificial Analysis Intelligence Index v5, scoring 59 points—one more than Claude Opus 5.5 and two more than GPT-6 Astra. The independent benchmark ranking is a narrow lead across nine tests, not a finding that Carbon is best at every task, and Google says the model is not yet generally available.
The index gives Carbon a 59-point total, compared with 58 for Claude Opus 5.5 at its maximum reasoning setting and 57 for GPT-6 Astra at maximum. GPT-6.1 Sol scores 55, while Google’s previous model, Gemini 4 Argon, scores 50. The source account says Carbon places first in four of the nine tests and falls behind competitors in the other five.
Carbon’s reported wins include HLE-Diamond, a difficult reasoning and knowledge test, where it scores 69% against Claude Opus 5.5’s 68%; Terminal-Bench-Science, at 65% against GPT-6 Astra’s 63%; and AutomationBench-AA, a business-workflow test where it narrowly exceeds Gemini 4 Argon. It also leads AA-Omniscience, a test that penalizes incorrect answers. The source says Carbon scores 54 on that measure, six points ahead of Claude Opus 5.5.
That last result does not mean Carbon answers the most questions correctly. The supplied figures put its plain accuracy at 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%. Artificial Analysis’ reported hallucination rate for Carbon is 11%, compared with about 50% for Claude Opus 5.5 and GPT-6 Astra. On a scoring system that penalizes wrong answers, abstaining rather than guessing can improve a model’s result.
Carbon also trails in several practical areas. In AA-Briefcase, which tests office-style knowledge work, it scores 1,609 Elo, below Claude Sonnet 5.5 at 1,812 and Claude Opus 5.5 at 1,800. In FrontierCode 2.0, its reported score is 41%, against 46% for Claude Opus 5.5. The source also reports losses on a physics-research test, a terminal-task benchmark and document reasoning.
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 Winner
For organizations comparing AI systems, the one-point margin is a reason to examine the underlying tasks rather than choose by leaderboard position alone. The index combines performance across different kinds of work; a company’s own use may place much more weight on coding, documents, research or software workflows than the index does. Carbon’s reported weaknesses in office-style work and real-world coding could matter more than its overall rank for some buyers.
Cost and access also affect the choice. The supplied account puts Carbon’s estimated cost at $7.76 per index task at standard prices: about 23% below Claude Opus 5.5 at $10.04, but around 1.2 times GPT-6 Astra at $6.24 and roughly six times GPT-6.1 Sol at $1.27. These are benchmark-task estimates, not a complete forecast of a company’s total operating costs. The source also lists introductory API pricing of $2 per million input tokens and $10 per million output tokens, later rising to $4 and $20, respectively.
For everyday users, there is no immediate change while general access remains unavailable. If the model becomes broadly available, the source expects its advantages to be more visible on long, demanding tasks than in routine chat. That is an interpretation, not a demonstrated outcome for all users.
Products Worth Considering
Anker SOLIX S2000 Portable Power Station, 2,010Wh, 1,500W
Bose TV Speaker (2nd Gen) Compact Soundbar with Bluetooth and HDMI
Bose TV Speaker Compact Soundbar with Bluetooth and HDMI-ARC, Black
As an affiliate, we earn on qualifying purchases.
How the New Index Scores Models
Artificial Analysis introduced version 5 on October 27, rebuilding the leaderboard around nine evaluations grouped into four equally weighted blocks: knowledge work, coding, science and general capability. The supplied breakdown assigns 25% of the total to each block, with individual tests carrying different weights within them. Several evaluations involve longer agent tasks, such as working in a terminal, producing reports or changing code.
According to the source, 60% of v5 uses private test sets or held-out answers, intended to reduce the chance that models benefit simply from memorizing public questions. The ranking nevertheless measures performance on its chosen tests, for the settings tested. A different benchmark mix or model configuration could produce a different order; the index is not a general verdict on which system is “the smartest.”
The source describes the release as a rapid leaderboard shift: Claude Opus 5.5 topped the rebuilt index on Tuesday, then Carbon took the lead the next day. The supplied material does not provide independent documentation beyond its account of the rankings, so individual scores and pricing figures here should be understood as reported by that source and attributed to Artificial Analysis where applicable.
Access and Real-World Results Remain Open
Google has not provided a public release date in the supplied material. The stated rollout order is a subset of cyber defenders, then trusted partners, followed by developers, businesses and consumers, but timing and eligibility for those stages are unspecified. It is also unclear when the listed introductory API prices would apply to a wider customer base.
The overall index margin is only one point, and the source does not provide uncertainty ranges or enough detail to determine how sensitive that difference is to evaluation variation. The reported scores also do not establish how Carbon will perform on a particular organization’s data, tools or workflows. Users should not read the benchmark’s hallucination measure as a guarantee that the model will avoid errors in other settings.
Watch for Broader Access and Testing
The next stated steps are Google’s rollout from selected cyber defenders to trusted partners and, later, developers, businesses and consumers. No dates have been announced for those stages in the material provided. Broader access would allow more users to assess latency, task quality, reliability and cost outside the benchmark setting.
For organizations, the practical next step is to compare models on representative tasks and measure cost per completed task, not just cost per benchmark run or overall rank. Until Google expands availability and customers can test Carbon directly, its top position remains a reported result on Artificial Analysis’ index rather than evidence of a settled winner for general use.
Key Questions
What score did Gemini 4 Carbon receive?
It scored 59 points on the 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 best at every task?
No. The supplied results say Carbon leads four of nine tests and loses the other five, including office-style knowledge work and real-world coding.
Why can Carbon lead an uncertainty-penalized test despite lower accuracy?
The source reports that Carbon answers fewer questions correctly than some rivals on plain accuracy, but has a lower reported hallucination rate. A test that penalizes wrong answers can reward a model for declining to guess.
Can the public use Gemini 4 Carbon now?
Not according to the supplied account. Google is first giving access to a subset of cyber defenders, with wider rollout stages described but no public dates.
Should companies switch to Carbon based on this score?
The ranking alone does not settle that decision. Companies should test models on their own workflows and compare quality, reliability, availability and the cost of completed work.
Source: ThorstenMeyerAI.com
