🔍 Read the full analysis: Gemini 4 Carbon’s #1 AI Model Ranking: A Small Lead, Bigger Questions on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Gemini 4 Carbon ranks first in Artificial Analysis’s Intelligence Index v5 with 59 points, one ahead of Claude Opus 5.5 and two ahead of GPT-6 Astra. It leads four of the nine tests, and the benchmark’s overall ranking does not mean it is best at every task. Google says Carbon is not publicly available yet and is starting with a small group of cyber defenders.
Google’s Gemini 4 Carbon ranks first in the Artificial Analysis Intelligence Index v5 with 59 points, one point ahead of Claude Opus 5.5 and two ahead of GPT-6 Astra. The result puts Carbon at the top of this particular nine-test benchmark, but it does not establish that the model is best for every task—and Google has not made it publicly available.
Artificial Analysis rebuilt its leaderboard on October 27, introducing version 5 of the index. Its scores place Gemini 4 Carbon, set to high reasoning, at 59 points; 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, scores 50. The ranking is a combined result across nine tests, not a measure of performance on every possible use.
Carbon comes first on four of the nine tests: HLE-Diamond, Terminal-Bench-Science, AA-Omniscience and AutomationBench-AA. It does not lead the other five. The supplied results put Claude models ahead on office-style knowledge work, Claude Opus 5.5 ahead on real-world coding, GPT-6 Astra ahead on physics research, Kimi K3.1 Pro ahead on terminal tasks, and Meta’s Muse Wonder Full ahead on document reasoning.
The benchmark’s reported cost per index task is $7.76 at standard prices. That is below Claude Opus 5.5’s reported $10.04, but above GPT-6 Astra’s $6.24 and GPT-6.1 Sol’s $1.27. The source also lists an introductory API price of $2 per million input tokens and $10 per million output tokens, later rising to $4 and $20 respectively. These prices are not the same as a guaranteed cost for a company’s own workload.
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 Win
The ranking may influence which models businesses test, but the small gap at the top makes the task-level results and operating cost more useful than the headline position alone. A company doing science research or business workflows may value Carbon’s reported strengths; an organization focused on coding or office-document work may find a rival performs better on the relevant benchmark.
Artificial Analysis’s v5 gives equal weight to four broad categories: knowledge work, coding, science and general performance. Each category contributes 25% to the total, while individual tests have different weights within those blocks. That design makes a single overall score convenient for comparison, but it can obscure differences in what a model is good at—and the cost of using it.
For buyers, the practical question is not simply which model holds first place. It is whether the model completes their own tasks accurately and reliably at an acceptable cost per finished task. The benchmark’s task prices and scores are reference points; they do not establish results for a particular company’s software, data or workflow.
Popular right now
KitchenAid Artisan Series 5-Quart Tilt-Head Stand Mixer, Dried Rose (KSM150PS)
Samsung 48-Inch S90H OLED 4K Smart TV (2026)
KitchenAid Artisan 5-Quart Tilt-Head Stand Mixer, Almond Cream
As an affiliate, we earn on qualifying purchases.
How Artificial Analysis Built Version 5
Artificial Analysis describes its Intelligence Index as a combined score from nine tests. Version 5 divides them among knowledge work, coding, science and general capability, each with equal category weight. The listed tests include tasks involving reports, business software, coding agents, terminal-based science workflows, and questions that reward models for not guessing when uncertain.
The source says 60% of the index uses private test sets or held-out answers, intended to limit the effect of models having memorized test questions. Many evaluations also involve longer agent-style tasks rather than short chat prompts. Those features make the benchmark more focused on extended work, but the results still apply to the specified tests, model versions and settings.
Carbon’s Omniscience result illustrates how test design affects rankings. It scores 54 on the benchmark’s penalty-adjusted measure, while its reported plain accuracy is 58%. Claude Fable 5.1, Claude Opus 5.5, GPT-6.1 Sol and GPT-6 Astra post higher plain-accuracy figures in the supplied comparison. Artificial Analysis reports an 11% hallucination rate for Carbon on this measure, versus about 50% for Claude Opus 5.5 and GPT-6 Astra. In this test, fewer wrong answers can matter more than answering more questions correctly.
Availability and Real-World Results
Google has not given a public release date for Gemini 4 Carbon. The reported rollout plan proceeds from a subset of cyber defenders to trusted partners, then to developers, businesses and consumers, but the source does not specify dates or eligibility criteria for those stages.
The index also cannot settle how Carbon will compare on a particular customer’s work. The reported scores concern defined tests and settings; they do not show whether a business will save time or money after accounting for its own tools, review requirements and usage volume. The source material gives the task-level wins and losses, but does not provide full results for every test or explain all benchmark methodology in detail.
The source’s account of a separate coding-agent comparison is incomplete: it reports scores of 62 for Claude Code with Claude Opus 5.5 and 59 for Google’s Antigravity CLI with Carbon, then ends mid-sentence while attributing a comment to Artificial Analysis. That truncated passage does not support a fuller claim about the company’s assessment.
Google’s Staged Release Plan
Google’s stated sequence is to begin with a small group of cyber defenders, then extend access to trusted partners, developers, businesses and consumers. The source gives no timetable for those steps. Until public access begins, everyday users cannot independently try Carbon, and companies outside the initial groups may not be able to evaluate it directly.
As access expands, the key developments to watch are Google’s release dates and API terms, independent testing beyond the nine index measures, and results on specific business workloads. For organizations considering a switch, the reported data supports comparing models on representative tasks and tracking both accuracy and cost per completed job, rather than relying on the overall rank alone.
Key Questions
What does Gemini 4 Carbon’s first-place ranking mean?
It means Carbon scored 59 points in Artificial Analysis’s Intelligence Index v5, one point above Claude Opus 5.5. The index combines nine tests; it does not show that Carbon is best for every use.
Is Gemini 4 Carbon available to the public?
No, according to the source material. Google is starting with a subset of cyber defenders and plans to expand access in stages, but has not provided a public release date.
Which tasks does Carbon lead?
The supplied results show Carbon leading on HLE-Diamond, Terminal-Bench-Science, AA-Omniscience and AutomationBench-AA. It trails other models on several coding, office-work, terminal and document-reasoning tests.
How much does Carbon cost to use?
The source reports a benchmark cost of $7.76 per index task at standard prices. Its listed introductory API rates are $2 per million input tokens and $10 per million output tokens, later rising to $4 and $20. Actual costs depend on use and are not established by the index.
Does Carbon answer more knowledge questions correctly than its rivals?
Not in the plain-accuracy comparison provided. Several Claude and GPT models score higher on correct answers, while Carbon ranks first on AA-Omniscience because the test also penalizes wrong answers and its reported hallucination rate is lower.
Source: ThorstenMeyerAI.com
