🔍 Read the full analysis: What To Know About Gemini 4 Carbon’s Narrow AI Model Lead 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. The result reflects performance across nine tests, not a universal ranking: Carbon leads four and trails on five, costs more than some rivals per index task, and is not yet generally available.
Google’s Gemini 4 Carbon has taken the lead 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 ranking gives Carbon the highest combined score across nine tests, but it does not establish that it is best for every task—and Google has not yet made it publicly available.
Artificial Analysis launched version 5 of its leaderboard on October 27, rebuilding the index around nine tests divided into four equally weighted areas: knowledge work, coding, science and general capability. In the published results, Claude Opus 5.5 scored 58, GPT-6 Astra 57, GPT-6.1 Sol 55 and Google’s earlier Gemini 4 Argon 50. The Carbon result is for its high-reasoning setting; the rival scores cited are at their specified maximum settings.
Carbon ranked first on four of the nine tests: HLE-Diamond, Terminal-Bench-Science, AA-Omniscience and AutomationBench-AA. Its results included 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, which tests business workflows across software tools, Carbon scored 78%, slightly ahead of Gemini 4 Argon.
The model placed behind competitors on the other five tests. In office-style knowledge work, Claude Sonnet 5.5 and Claude Opus 5.5 scored higher. Claude Opus 5.5 also led on real-world coding, while GPT-6 Astra led on physics research, Kimi K3.1 Pro on terminal tasks and Meta’s Muse Wonder Full on document reasoning. The supplied results put Carbon at 41% on FrontierCode 2.0, against 46% for Claude Opus 5.5.
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 With Trade-Offs
The score matters because the index covers work that can involve long, tool-using tasks, including research workflows, software development and business processes. A strong result may make Carbon a candidate for organizations evaluating models for those uses. But its narrow overall lead and uneven test results mean the ranking alone is not a sound basis for replacing another model.
Cost and access also affect practical choices. Artificial Analysis lists Carbon at $7.76 per index task at standard prices. That is 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. These are index-task estimates, not a guaranteed price for a company’s own workload. The source also lists introductory API pricing of $2 per million input tokens and $10 per million output tokens, later rising to $4 and $20 respectively.
For everyday users, the immediate effect is limited: Carbon is not publicly available. For businesses, the useful question is not simply which model ranks first, but which performs reliably on their actual tasks at an acceptable cost.
Products Worth Considering
Samsung Galaxy Tab S10 FE 128GB Wi-Fi Android Tablet, Blue
Amazon Kindle Paperwhite Signature Edition (12th Generation, 2024), 32 GB, Metallic Black
Apple AirPods Pro 3 Wireless Earbuds
As an affiliate, we earn on qualifying purchases.
How the New Index Scores Models
Artificial Analysis describes the index as a combined score from nine benchmarks, with knowledge work, coding, science and general tests each accounting for 25% of the total. Within those categories, individual tests have different weights. For example, FrontierCode 2.0 accounts for 15% of the full index, as do AA-Briefcase and Terminal-Bench-Science.
Version 5 uses private test sets or held-out answers for 60% of its evaluation, according to the source material. Many tasks require models to act as agents—for example, working in a terminal, preparing reports or fixing code. That design aims to test more than short question-and-answer performance, but the result remains specific to the index’s selected tasks, test design and model settings.
One result illustrates how a benchmark’s scoring rules can change the ranking. Carbon scored 58% on plain accuracy in AA-Omniscience, below several Claude and GPT models. Yet it led the test’s score because the evaluation penalizes wrong answers and Carbon had an 11% hallucination rate, as defined by Artificial Analysis. Claude Opus 5.5 and GPT-6 Astra were each at about 50% on that measure. The index therefore rewards restraint as well as correct answers; it does not show that Carbon knows more than those models.
Access and Real-World Performance
Google has not given a date for general availability. The stated rollout order is a subset of cyber defenders, followed by trusted partners, then developers, businesses and consumers. The source material does not specify how large each group will be, what eligibility rules will apply or when each stage will begin.
The published ranking also cannot establish how Carbon will perform across every company’s software, data and workflows. The supplied account notes that a separate coding-agent comparison scored Claude Code with Claude Opus 5.5 at 62 and Google’s Antigravity CLI with Carbon at 59, but it provides no further detail about the evaluation or the benchmark’s full methodology. Carbon’s measured advantage on the overall index is just one point, and changes in tasks, settings or scoring could alter the order.
Google’s Staged Rollout Ahead
Google’s next stated step is to provide Carbon to a subset of cyber defenders, then expand access to trusted partners. The company has not announced dates for those stages or for release to developers, businesses and consumers. Until broader access begins, users cannot independently test the model through a public release.
Organizations weighing a future adoption can compare Carbon with alternatives on their own work, including accuracy, refusal behavior, completion time and cost per finished task. The benchmark provides a comparison point, not a substitute for those evaluations. The next clear developments to watch are Google’s access timetable, final API pricing and performance evidence from users outside the initial group.
Key Questions
What is Gemini 4 Carbon’s score on the index?
Carbon scored 59 points on Artificial Analysis Intelligence Index v5. Claude Opus 5.5 scored 58, and GPT-6 Astra scored 57.
Does the ranking mean Carbon is the best AI model for every task?
No. The index combines results from nine selected tests. Carbon placed first on four and behind other models on five, so its overall lead does not mean it leads every use case.
Can the public use Gemini 4 Carbon now?
No public release date has been announced. Google says it is starting with a subset of cyber defenders, then plans to expand access to trusted partners, developers, businesses and consumers.
How much does Carbon cost?
The source lists an estimated $7.76 per index task at standard prices. It also gives introductory API prices of $2 per million input tokens and $10 per million output tokens, later rising to $4 and $20. Actual costs can vary with usage.
Source: ThorstenMeyerAI.com
