🔍 Read the full analysis: Gemini 4 Carbon Leads AI Models By One Point: How To Read The Ranking 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 scores 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 four of the nine tests, but the index is a weighted benchmark rather than a universal measure of model quality, and Carbon is not publicly available yet.
Google’s Gemini 4 Carbon ranks first on 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 nine-test benchmark, not a finding that Carbon is best at every kind of AI work; Google has also not made the model publicly available.
The index combines results from nine tests grouped into four equally weighted areas: knowledge work, coding, science and general performance. Carbon took first place in four individual tests, according to the source report: hard reasoning and knowledge questions, science workflows conducted in a terminal, a knowledge test that penalizes incorrect answers, and business workflows using software tools.
It did not lead the other five tests. Claude models performed better on office-style knowledge work, Claude Opus 5.5 led real-world coding, GPT-6 Astra led a physics research test, Kimi K3.1 Pro led terminal tasks, and Meta’s Muse Wonder Full led document reasoning. The overall score compresses those different results into one number, so a one-point advantage does not mean Carbon wins across the board.
The report puts Carbon’s cost at $7.76 per index task at standard prices. That is about 23% below Claude Opus 5.5 at $10.04, but above GPT-6 Astra at $6.24 and GPT-6.1 Sol at $1.27. The source says Google plans an introductory API price of $2 per million input tokens and $10 per million output tokens, later rising to $4 and $20; the relationship between those token prices and the reported per-task benchmark cost depends on usage.
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
What a One-Point Lead Shows
For businesses comparing models, the result is a reason to examine Carbon, not an automatic reason to switch. Task-level performance and cost per completed job may matter more than the overall rank, particularly when a workflow depends on coding, office documents or predictable answers. The benchmark’s reported test-specific results show that another model can be a stronger choice for particular work.
The standout trade-off is in knowledge questions. The source report says Carbon answered 58% correctly on AA-Omniscience, below several Claude and GPT models, but scored higher on that test because it gave fewer wrong answers: its reported hallucination rate was 11%, compared with about 50% for Claude Opus 5.5 and GPT-6 Astra. That distinction matters where unsupported answers carry a cost. It does not mean Carbon knows more, or that its low error rate will carry over to every real-world use.
For ordinary users, there is no immediate change in access. Google is starting with a limited group of cyber defenders, according to the source, before a planned rollout to trusted partners and then wider groups. No public release date is given.
Products Worth Considering
Amazon Kindle Scribe Colorsoft 64GB (1st Generation)
Jackery HomePower 3000 Portable Power Station with Two 200W Solar Panels
Anker SOLIX S2000 Portable Power Station, 2,010Wh, 1,500W
As an affiliate, we earn on qualifying purchases.
How the Index Builds Its Score
Artificial Analysis launched version 5 on October 27, one day before the source report described Carbon taking the lead. The index assigns equal weight to four categories: knowledge work, coding, science and general tasks. Within those blocks, tests have different weights, so a model’s final score reflects both its performance and how much each test counts.
The source says 60% of the v5 tests use private test sets or held-out answers, intended to reduce the effect of models memorizing published questions. Many evaluations involve extended agent tasks, such as working in a terminal, producing reports or fixing code. Those tasks can offer a closer proxy for multi-step work than short question-and-answer tests, but the results still describe performance on the selected tasks and settings.
The ranking also shows how benchmark changes can alter the order. The source reports that Google’s earlier Gemini 4 Argon scored 50 on v5, while Carbon scored 59. That comparison is within this index and does not by itself establish how the models perform across other benchmarks, products or user workloads.
““number one” on one index is a narrower claim than it sounds.”
— Thorsten Meyer, ThorstenMeyerAI.com
Limits Behind the Ranking
The supplied material does not include the full underlying benchmark report, methodology tables or independent replication. It also does not provide the full text of Artificial Analysis’s comment on the separate Coding Agent Index: the source excerpt ends mid-sentence. That comparison is reported as 62 for Claude Code with Claude Opus 5.5 and 59 for Google’s Antigravity CLI with Carbon, but the complete explanation is not available here.
It remains unclear when Carbon will reach general users, how quickly Google’s staged rollout will proceed, and whether the proposed API prices will apply uniformly across access tiers. The index score also cannot settle how Carbon will perform on a company’s own data or workflows. Its cost per benchmark task should not be treated as a fixed price for every deployment.
Access and Independent Testing
According to the source report, Google plans to offer Carbon first to a subset of cyber defenders, then to trusted partners, developers, businesses and consumers. The company has not provided a timetable in the material supplied. Wider availability and published API terms would give prospective customers more information about access, price and practical use.
For now, organizations considering the model can compare it against alternatives on representative tasks and track both output quality and cost per completed task. Further benchmark results, the complete methodology and independent testing would help clarify whether Carbon’s narrow index lead translates into an advantage beyond the nine evaluations.
Key Questions
What is Gemini 4 Carbon’s score?
It scores 59 points on the Artificial Analysis Intelligence Index v5, one point ahead of Claude Opus 5.5 at 58.
Does the ranking mean Carbon is the best AI model?
No. It ranks first on this particular nine-test index. Carbon leads four tests and trails other models on five, so the ranking does not establish that it is best for every task.
Is Gemini 4 Carbon publicly available?
Not according to the source report. Google is beginning with a limited group of cyber defenders and has not announced a date for general access.
Why does Carbon lead the knowledge test if others answer more correctly?
The reported AA-Omniscience score penalizes wrong answers. Carbon’s lower reported hallucination rate helped it score well even though several competitors had higher raw accuracy.
How much does Carbon cost?
The source reports $7.76 per index task at standard prices and proposed introductory API rates of $2 per million input tokens and $10 per million output tokens. These are different pricing measures and do not establish the cost of every real-world task.
Source: ThorstenMeyerAI.com
