AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: Gemini 4 Carbon Is The New #1 AI Model—A Look Beyond The Score on ThorstenMeyerAI.com

Before you orderOffer from Amazon

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
Start your free Prime trial Free trial for eligible customers · Cancel anytime
As an affiliate, we earn on qualifying purchases.

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. It leads on four of nine tests, but trails rivals on others, costs more than some alternatives and is not yet publicly available.

Google’s Gemini 4 Carbon has taken the top spot in the 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 a universal measure of AI capability: Carbon ranks first on four of nine tests, and Google says access is initially limited to a small group of cyber defenders.

Artificial Analysis launched version 5 of its index on October 27, then published the new ranking the following day, according to the source report. The leaderboard scores models across nine tests, with four categories—knowledge work, coding, science and general capability—each contributing 25% of the total. The reported scores are 59 for Gemini 4 Carbon at high reasoning, 58 for Claude Opus 5.5 at max, 57 for GPT-6 Astra at max, 55 for GPT-6.1 Sol at max and 50 for Google’s Gemini 4 Argon.

Carbon leads on HLE-Diamond, a hard reasoning and knowledge test; Terminal-Bench-Science; AA-Omniscience; and AutomationBench-AA, which tests business workflows. Its results include 69% on HLE-Diamond, compared with 68% for Claude Opus 5.5, and 65% on Terminal-Bench-Science, compared with 63% for GPT-6 Astra. The overall index lead is just one point, and the model does not lead every component.

Its weaknesses include office-style knowledge work and real-world coding. On AA-Briefcase, Carbon scored 1,609 Elo, below Claude Sonnet 5.5 at 1,812 and Claude Opus 5.5 at 1,800. On FrontierCode 2.0, it scored 41%, compared with 46% for Claude Opus 5.5. The report also says GPT-6 Astra leads a physics research test, Kimi K3.1 Pro leads a terminal task test, and Meta’s Muse Wonder Full leads document reasoning.

At a glance
reportWhen: Published October 29, 2026; the index l…
The developmentArtificial Analysis’s new index ranks Google’s Gemini 4 Carbon first with 59 points, narrowly ahead of leading Claude and GPT models.
Thorsten Meyer AI
Index v5

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.

59Gemini 4 Carbon on Intelligence Index v5
1point ahead of Claude Opus 5.5 (58)
4 / 9tests where Carbon scores highest
60%of the index runs on private tests or held-out answers

The nine tests behind the score

Carbon at its highest setting (high) against the best model on each test.

TestWeightCarbonLeader 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 54Carbon 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 EloBehind 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 With Real Trade-Offs

The ranking matters because companies and developers use benchmarks to compare systems before testing them in their own products. But Carbon’s one-point advantage does not settle which model is best for a particular job. The index combines a defined set of tasks and model settings; a different workload or scoring method could produce a different order.

Price and availability also shape the practical choice. The source report estimates Carbon’s cost at $7.76 per index task at standard prices: about 23% below Claude Opus 5.5 at $10.04, but above GPT-6 Astra at $6.24 and far above GPT-6.1 Sol at $1.27. These are costs for the index’s tasks, not a universal price for every use. The report lists introductory API rates of $2 per million input tokens and $10 per million output tokens, rising later to $4 and $20 respectively.

For organizations, the results point to evaluating cost per completed task and performance on their own work rather than selecting a system by leaderboard position alone. For everyday users, the immediate effect is limited: Carbon is not yet generally available, and the source report says Google is beginning with selected cyber defenders.

Products Worth Considering

Kindle Paperwhite Signature Edition 32GB (2024 Model) - Metallic Black

Kindle Paperwhite Signature Edition 32GB (2024 Model) – Metallic Black

DJI Mini 4K Fly More Combo Drone with 4K Camera and 3 Batteries

DJI Mini 4K Fly More Combo Drone with 4K Camera and 3 Batteries

Bose QuietComfort True Wireless Earbuds with Active Noise Cancellation, Bluetooth 5.3, USB-C Charging, IPX4, Black

Bose QuietComfort True Wireless Earbuds with Active Noise Cancellation, Bluetooth 5.3, USB-C Charging, IPX4, Black

As an affiliate, we earn on qualifying purchases.

How Artificial Analysis Scores Models

Artificial Analysis describes its index as a combined score across nine evaluations, rather than a direct verdict on which system is “the smartest.” Version 5 gives equal weight to four blocks: knowledge work, coding, science and general tasks. Within those blocks, individual tests have different weights. For example, AA-Briefcase and FrontierCode 2.0 each account for 15% of the total, as does Terminal-Bench-Science.

The test design includes long, agent-style assignments: models may work in a terminal, produce reports or modify code. The source report says 60% of the index uses private test sets or held-out answers, intended to reduce the chance that systems have simply memorized questions. That design makes the scores relevant to some practical work, but results still depend on the selected tests, task weights and model configurations.

One of Carbon’s most distinctive results is on AA-Omniscience, which penalizes incorrect answers. Carbon scored 54 points, six ahead of Claude Opus 5.5, although the report says Carbon’s plain accuracy was 58%, below several Claude and GPT models. Its reported hallucination rate on that evaluation was 11%, compared with about 50% for Claude Opus 5.5 and GPT-6 Astra. The metric reflects the test’s treatment of wrong answers; it does not establish that Carbon is more accurate across all tasks.

Access and Benchmark Limits

Google has not provided a public release date in the supplied material. It says access will begin with selected cyber defenders, then extend to trusted partners, developers, businesses and consumers, but the timing and criteria for each stage are not specified.

The source material also ends during a sentence about Artificial Analysis’s assessment of Google’s coding tools, so that comment cannot be reported in full. More broadly, the reported index scores do not show how Carbon will perform across every real-world task, nor do they establish that the differences between close scores will be meaningful for a particular user. The pricing figures and ranking are those provided in the source report and may change.

Google’s Staged Release Plans

Google’s stated plan is to release Carbon in stages, starting with a subset of cyber defenders and then moving to trusted partners, developers, businesses and consumers. No dates were supplied for those stages. Until broader access begins, prospective users will not be able to check the benchmark results against their own workflows.

For now, the next useful developments are clearer access dates, final pricing and independent results on tasks beyond the nine index tests. Organizations considering the model can compare it with alternatives using their own workloads and account for both output quality and the cost of completing a task. The leaderboard may change as Artificial Analysis publishes further evaluations or updates its test set.

Key Questions

What is Gemini 4 Carbon’s score?

It scored 59 points on Artificial Analysis’s Intelligence Index v5, one point ahead of Claude Opus 5.5 and two ahead of GPT-6 Astra.

Does the index rank Carbon first on every test?

No. The source report says Carbon leads on four of nine tests. It trails leading Claude models on office-style work and real-world coding, among other areas.

Can the public use Gemini 4 Carbon now?

Not according to the supplied report. Google is beginning with a subset of cyber defenders and has not announced dates for wider access.

How much does Carbon cost?

The report estimates $7.76 per index task at standard prices. It lists introductory API rates of $2 per million input tokens and $10 per million output tokens, later increasing to $4 and $20. Actual costs depend on usage.

Does Carbon’s score mean it is the best model for every user?

No. The score summarizes performance on nine selected tests and depends on their weights and model settings. Users and organizations need to compare systems on the tasks they actually perform.

Source: ThorstenMeyerAI.com

You May Also Like

macOS Golden Gate Is A Buggy Mess

A macOS 27 ‘Golden Gate’ review documents menu bar crashes, UI glitches and a system-hanging eyedropper bug, despite Apple’s stability-focused pitch.

GPT-6 Astra’s Review Shows What Green Tests Missed In Claude’s Edit

An independent GPT-6 Astra review found reproducible benchmark and weight-conversion faults in a Claude-led model-editing experiment.

The First Nintendo Switch 2 Emulator For PC Has Just Been Released

A new Eden-NX2 build can boot Mario Kart World and Donkey Kong: Bananza, but performance is poor and the emulator remains experimental.

What The Records Reveal About Nvidia H200 Servers And A US-Blacklisted School

Archived notices show Northwestern Polytechnical University’s cybersecurity school rented servers with eight Nvidia H200 GPUs; licensing details are unknown.