🔍 Read the full analysis: Gemini 4 Carbon And The AI Model Ranking: Beyond The One-Point Lead on ThorstenMeyerAI.com
Get the little things that make your day delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
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, and Google has not made it publicly available. The ranking is one benchmark result, not a universal measure of model quality.
Google’s Gemini 4 Carbon ranks first on the newly rebuilt Artificial Analysis Intelligence Index v5, scoring 59 points—one ahead of Claude Opus 5.5 and two ahead of GPT-6 Astra. The result places Carbon at the top of this particular nine-test benchmark, but does not establish it as the best model for every task; Google is also limiting access while it begins a staged rollout.
The ranking gives Claude Opus 5.5 a score of 58, GPT-6 Astra 57, GPT-6.1 Sol 55 and Google’s previous Gemini 4 Argon 50. Carbon records the highest score in four of the nine underlying tests, including hard reasoning and knowledge questions, science workflows carried out in a terminal, business workflows across software tools, and a test that rewards models for avoiding unsupported answers.
Its leads are not uniform. Carbon scores 69% on HLE-Diamond, compared with 68% for Claude Opus 5.5, and 65% on Terminal-Bench-Science, against 63% for GPT-6 Astra. It also leads AA-Omniscience, with 54 points. However, the source reports that Carbon answers fewer questions correctly in that test than several rivals: its plain accuracy is 58%, compared with 66% for Claude Opus 5.5 and 62% for both GPT-6 Astra and GPT-6.1 Sol.
Artificial Analysis’s result reflects a penalty for wrong answers: the source gives Carbon an 11% hallucination rate, compared with about 50% for Claude Opus 5.5 and GPT-6 Astra. Carbon’s score on this test can therefore be higher even though other models have higher raw accuracy. It is not the lowest-hallucination model in the comparison: Gemini 4 Argon is reported at 4%.
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
For companies choosing a model, the result points to a task-by-task trade-off, not a simple winner. Carbon is strong on some reasoning, science and business workflow tests, while other systems score higher on office-style work, real-world coding, physics research, terminal tasks and document reasoning. A company’s own workload may not resemble the benchmark’s mix of tasks.
Price and speed also affect the practical choice. The source puts Carbon’s cost at $7.76 per index task at standard prices: about 23% below Claude Opus 5.5 at $10.04, but roughly 1.2 times GPT-6 Astra at $6.24 and about six times GPT-6.1 Sol at $1.27. Carbon is described as slow, though the supplied material does not provide a latency figure. Buyers will need to judge cost and completion time against the value of accurate results on their own tasks.
For everyday users, the ranking has no immediate practical effect because Carbon is not publicly available. If it reaches consumer products, the source suggests its differences may be most apparent on lengthy, difficult work rather than routine chat; that is an expectation, not a confirmed user experience.
Products Worth Considering
Google Nest Learning Thermostat (4th Gen) with Nest Temperature Sensor (2nd Gen), Obsidian
Philips Hue Secure Wired Video Doorbell, 2K, Black
Roborock Qrevo CurvX Robot Vacuum and Mop
As an affiliate, we earn on qualifying purchases.
How Artificial Analysis Built the Score
Artificial Analysis launched Index v5 on October 27, replacing its previous leaderboard with a composite based on nine tests. The tests are divided into four equally weighted areas: knowledge work, coding, science and general ability. Individual test weights vary within those blocks; for example, AA-Briefcase and FrontierCode 2.0 each account for 15% of the full index.
The test set covers work beyond short question-and-answer exchanges. Models may act as agents in a terminal, produce reports or modify code. The source says 60% of v5 uses private test sets or held-out answers, intended to reduce the benefit of memorizing known questions. Even so, the index measures performance on its own selected tests and model settings. Different tasks, weights or settings could produce a different ordering.
Carbon’s reported weaknesses illustrate that limitation. On AA-Briefcase, an office-style knowledge-work test worth 15% of the index, it scores 1,609 Elo, below Claude Sonnet 5.5 at 1,812 and Claude Opus 5.5 at 1,800. On FrontierCode 2.0, it scores 41%, against 46% for Claude Opus 5.5. The source also says the Claude Code and Antigravity CLI tool-model combinations score 62 and 59, respectively, on a separate Coding Agent Index.
Access, Timing and Real-World Results
Google has not provided a public release date in the supplied material. Its stated sequence is a limited rollout to cyber defenders, then trusted partners, followed by developers, businesses and consumers. The size of each group, the pace of expansion and the criteria for access are not specified.
The source gives introductory API pricing of $2 per million input tokens and $10 per million output tokens, rising later to $4 and $20, but does not specify when that change takes effect. It also does not provide a detailed latency measurement or enough information here to compare total costs across different workloads. Benchmark performance may not predict results on a particular company’s systems, prompts or tools.
A Staged Rollout and Buyer Testing
The next milestone is Google’s expansion beyond the initial group of cyber defenders to trusted partners and then other users. No calendar dates are given, so availability for developers, businesses and consumers remains uncertain.
Until access widens, organizations can use the benchmark as one input rather than a purchasing decision. The useful next step is to test Carbon and competing models on representative internal tasks, tracking accuracy, refusal or uncertainty behavior, speed and cost per completed task. Further Artificial Analysis results and real-world evaluations may show whether Carbon’s narrow composite lead holds up beyond the nine tests.
Key Questions
What score did Gemini 4 Carbon receive?
Carbon scored 59 points on Artificial Analysis’s Intelligence Index v5, one point above Claude Opus 5.5 at 58 and two above GPT-6 Astra at 57.
Does the ranking prove Carbon is the best AI model?
No. It places Carbon first on one nine-test index at a particular model setting. Carbon leads four tests and trails competitors on others, so performance depends on the task and evaluation method.
Can the public use Gemini 4 Carbon now?
Not according to the supplied report. Google is starting with a subset of cyber defenders, then plans to expand to trusted partners, developers, businesses and consumers. No public launch date is stated.
How much does Carbon cost?
The reported standard-price estimate is $7.76 per index task. 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; it does not give a date for that increase.
Source: ThorstenMeyerAI.com
