🔍 Read the full analysis: What The Gemini 4 Carbon AI Model Ranking Doesn’t Tell You on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
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 leads four of nine tests, but trails rivals on other tasks, costs more than some alternatives and is not yet publicly available. The ranking measures performance on a defined set of tests, not a universal measure of AI ability.
Google’s Gemini 4 Carbon scored 59 points to lead Artificial Analysis’s Intelligence Index v5, one point ahead of Claude Opus 5.5 and two ahead of GPT-6 Astra. The result puts Carbon first on this particular benchmark, but it does not establish that it is the best model for every task: Carbon won four of the index’s nine tests, and Google has not yet made it generally available.
The index combines nine tests into a single score, with four equally weighted categories: knowledge work, coding, science and general performance. Carbon placed first on HLE-Diamond, Terminal-Bench-Science, AA-Omniscience and AutomationBench-AA. It scored 69% on HLE-Diamond, one percentage point above Claude Opus 5.5, and 65% on Terminal-Bench-Science, two points above GPT-6 Astra.
It fell behind on other measures. Claude models led office-style knowledge work, Claude Opus 5.5 led the real-world coding test, GPT-6 Astra led a physics research test, Kimi K3.1 Pro led terminal tasks, and Meta’s Muse Wonder Full led document reasoning. In the index’s AA-Briefcase test, Carbon scored 1,609 Elo, compared with 1,812 for Claude Sonnet 5.5 and 1,800 for Claude Opus 5.5.
The reported standard cost to run the index tasks on Carbon is $7.76 per task. That is below Claude Opus 5.5’s $10.04, but above GPT-6 Astra’s $6.24 and well above GPT-6.1 Sol’s $1.27. Google’s introductory API pricing is listed as $2 per million input tokens and $10 per million output tokens, rising later to $4 and $20, respectively. Those prices and the benchmark’s task costs describe different measures and should not be treated as interchangeable.
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
For companies choosing a model, Carbon’s first-place score is a reason to test it, not a reason to assume it will outperform alternatives across their workloads. The index itself shows meaningful variation: the winner changes between coding, office work, research and document tasks. A model that scores well overall may still be a poor fit for a particular team’s software, documents or error tolerance.
The ranking also reflects cost and response behavior, not just how many questions a model answers correctly. On AA-Omniscience, Carbon scored 54 points and topped the test, while Claude Fable 5.1, Claude Opus 5.5 and GPT-6 variants had higher plain-answer accuracy. Artificial Analysis says Carbon’s lower rate of incorrect answers among questions it did not answer correctly helped it win a test that penalizes wrong answers. This distinction matters where an uncertain answer is safer than a confident mistake.
Access is another practical limit. Google says Carbon is initially being provided to a subset of cyber defenders, with a staged rollout planned for trusted partners and then developers, businesses and consumers. Without a public release date, most prospective users cannot evaluate it directly today. For current decisions, the ranking is evidence about benchmark performance, not a substitute for testing on real work.
Products Worth Considering

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

Apple AirPods Pro (2nd Generation) Wireless Earbuds with USB-C MagSafe Charging Case

Bose QuietComfort Ultra Earbuds (2nd Gen), Wireless Noise-Cancelling Earbuds, White Smoke
As an affiliate, we earn on qualifying purchases.
How Index v5 Builds Its Score
Artificial Analysis introduced version 5 on October 27, 2026, replacing its prior index structure. The nine tests are divided into four equally weighted blocks: knowledge work, coding, science and general performance. Within those blocks, individual tests carry different shares; for example, AA-Briefcase accounts for 15% of the total and AutomationBench-AA for 5%.
The index is designed to assess models on more than short question-and-answer prompts. Many evaluations involve longer agent-style tasks, such as using a terminal, producing reports or working with code. Artificial Analysis says 60% of v5 uses private test sets or held-out answers, which is intended to reduce the chance that models have memorized test questions. The source report says the longer tasks also make runs more expensive.
The headline scores place Gemini 4 Carbon at 59, Claude Opus 5.5 at 58, GPT-6 Astra at 57 and GPT-6.1 Sol at 55. Google’s previous model, Gemini 4 Argon, scored 50. Those figures show performance under the index’s chosen tests and settings; they are not a direct measure of every capability or of how the models will perform in a specific product.
Access and Real-World Performance
Google has not announced a date for general availability, and the timing of each stage in its planned rollout is unspecified. The source material also does not provide full details about how the reported API prices will apply across products, usage patterns or contract terms.
The supplied report is incomplete: it ends mid-sentence in its coding section after beginning to attribute a statement to Artificial Analysis. That missing wording cannot be reconstructed, so this account does not make a claim based on it. More broadly, the reported rankings do not show how Carbon performs across every possible workload, nor how results may change with different model settings, tools or benchmark choices.
Carbon’s 59-point total is close to the next two models, with gaps of only one and two points. The source material gives the scores but does not establish whether those differences are statistically meaningful. Readers should treat the order as the result reported for this benchmark run, rather than proof of a durable or universal advantage.
Watch for Rollout and Independent Tests
Google’s stated plan is to begin with a limited group of cyber defenders, then extend access to trusted partners, developers, businesses and consumers. The company has not supplied dates for those stages. A public release would allow more users to compare Carbon with alternatives on their own workloads and see how advertised token prices translate into actual costs.
For now, organizations evaluating AI systems can compare the task-level results with their own needs and test candidate models against representative work. The next useful developments will be clearer availability terms, independently reported results beyond the index’s nine tests, and evidence about cost and reliability in day-to-day use. Until then, Carbon’s ranking confirms a narrow benchmark lead, while its broader practical advantage remains unestablished.
Key Questions
Did Gemini 4 Carbon rank first overall?
Yes. It scored 59 points on Artificial Analysis Intelligence Index v5, ahead of Claude Opus 5.5 at 58 and GPT-6 Astra at 57. That is a first-place result on this index, not a universal ranking of AI models for every task.
What does the index measure?
It combines nine tests across knowledge work, coding, science and general performance. Several tests involve longer tasks, such as agent workflows, terminal work or producing reports.
Is Gemini 4 Carbon available to the public?
Not according to the source report. Google is starting with a subset of cyber defenders and plans further access stages, but has not given a public release date.
Does Carbon beat rivals on every test?
No. Carbon leads four of nine tests, while other models lead office-style knowledge work, real-world coding, physics research, terminal tasks and document reasoning.
How much does it cost to run the benchmark tasks?
The report lists a standard cost of $7.76 per index task for Carbon. That is a benchmark-run cost estimate, separate from Google’s quoted API rates per million tokens.
Source: ThorstenMeyerAI.com
