🔍 Read the full analysis: Gemini 4 Carbon Tops The AI Ranking By One Point—What Else Matters? on ThorstenMeyerAI.com
Prime made for students and young adults
- Fast, free delivery for dorm and study essentials
- Prime Video and Amazon Music included
- Member-only deals
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 on four of nine tests, but loses on five, costs more than some rivals per index task and is not publicly available yet.
Google’s Gemini 4 Carbon now leads the Artificial Analysis Intelligence Index v5 with 59 points, one ahead of Claude Opus 5.5 and two ahead of GPT-6 Astra. The ranking, updated a day after the benchmark’s rebuild, puts Carbon at the top of this particular nine-test index; it does not establish that the model is best for every task, and Google has not made it publicly available.
Artificial Analysis combines results from nine tests into a single score, with equal weight assigned to knowledge work, coding, science and general capabilities. Carbon ranks first on four tests: hard reasoning and knowledge questions, science workflows in a terminal, knowledge questions that penalize wrong answers, and business workflows across software tools. It does not lead the other five.
The overall margin is narrow: Claude Opus 5.5 scored 58, GPT-6 Astra 57 and GPT-6.1 Sol 55. Google’s previous model, Gemini 4 Argon, scored 50. On individual tests, Carbon scored 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. On AutomationBench-AA, it scored 78%, slightly above Argon.
Carbon’s result on AA-Omniscience reflects its willingness to leave questions unanswered rather than guess. It scored 58% on plain answer accuracy, below several Claude and GPT models, but had an 11% hallucination rate as defined by Artificial Analysis. The source says Claude Opus 5.5 and GPT-6 Astra were each at about 50% on that measure. Because this test penalizes incorrect answers, a lower rate of wrong responses can produce a stronger score even when a model answers fewer questions correctly overall.
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
The Score Does Not Settle Model Choice
The ranking offers a useful comparison, but its one-point gap is not a broad verdict on which system a company or individual should choose. Carbon trails Claude models on office-style knowledge work and real-world coding, GPT-6 Astra on physics research, Kimi K3.1 Pro on terminal tasks, and Meta’s Muse Wonder Full on document reasoning. Performance depends on the job being tested.
Price and access also matter. Artificial Analysis estimates $7.76 per index task for Carbon at standard prices. That is 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. Those figures make the cost of completing a particular task more relevant than the overall rank alone. Companies will need to test models on their own workloads before making a switch; everyday users cannot try Carbon yet.
Popular right now
Vitamix Propel Series 510 Professional-Grade Blender, 3 Programs, Black
Roborock Qrevo CurvX Robot Vacuum and Mop
LG C2 Series 65-Inch OLED evo 4K Smart TV (OLED65C2PUA, 2022)
As an affiliate, we earn on qualifying purchases.
How the New Index Is Built
Artificial Analysis launched version 5 on October 27 and rebuilt its leaderboard around nine tests. Each of four groups—knowledge work, coding, science and general tasks—accounts for 25% of the composite score. Tests include office and business workflows, coding agents, science tasks and difficult knowledge questions.
The index focuses on extended tasks, including agents working in terminals, preparing reports and fixing code. Artificial Analysis says 60% of v5 uses private test sets or held-out answers, intended to reduce the chance that models have memorized questions. The resulting score still reflects a particular set of tests and model settings. Change the tasks or their weights and the order may change.
Carbon is also behind on specific measures that matter to users. It scored 1,609 Elo on AA-Briefcase, compared with 1,812 for Claude Sonnet 5.5 and 1,800 for Claude Opus 5.5. On FrontierCode 2.0, it scored 41%, below Claude Opus 5.5’s 46%. In a separate coding-agent comparison described by the source, Claude Code with Opus 5.5 scored 62, while Google’s Antigravity CLI with Carbon scored 59.
Public Access and Real-World Fit
Google has given no date for wider access. The stated rollout order is a subset of cyber defenders, then trusted partners, developers, businesses and consumers. The source does not specify how large the first group is, which regions are included, or when each later stage will begin.
The benchmark also cannot establish how Carbon will perform across every company’s tools, data and workflows. Its aggregate score combines nine chosen tests, and several results depend on the model’s ability to abstain from answering. The reported task-cost estimate is not a guarantee of what every user will pay: actual costs depend on use, and the source describes introductory API rates that are set to change.
Google’s Rollout and Pricing
Google’s reported plan is to expand access from selected cyber defenders to trusted partners, followed by developers, businesses and consumers. No schedule has been announced, so the timing of those stages remains unknown. Until access expands, the ranking is a benchmark result rather than a model that most users can test themselves.
The source lists introductory API prices of $2 per million input tokens and $10 per million output tokens, later rising to $4 and $20, respectively. Companies comparing Carbon with competing models can assess those prices alongside performance on their own tasks and the cost of completing each task. The next meaningful evidence will come from broader availability and independent testing beyond the nine index benchmarks.
Key Questions
What is Gemini 4 Carbon’s score?
It scored 59 on Artificial Analysis Intelligence Index v5, one point ahead of Claude Opus 5.5 and two points ahead of GPT-6 Astra.
Does the ranking mean Carbon is the best model for every task?
No. Carbon led on four of nine tests and lost the other five. The results vary by task, including coding, office work, science and document reasoning.
Can the public use Gemini 4 Carbon now?
No. Google is starting with a subset of cyber defenders, then plans to expand to partners, developers, businesses and consumers. No dates for those stages have been given.
How much does Carbon cost to run?
Artificial Analysis estimates $7.76 per index task at standard prices. The source also lists introductory API rates of $2 per million input tokens and $10 per million output tokens, later rising to $4 and $20.
Why did Carbon lead the knowledge test if other models answered more correctly?
AA-Omniscience penalizes wrong answers. Carbon had lower plain accuracy than several rivals but a lower reported hallucination rate, so its more cautious answers earned a higher score on that test.
Source: ThorstenMeyerAI.com
