🔍 Read the full analysis: Gemini 4 Carbon At #1: The AI Ranking Details Worth A Second Look 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’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, costs more than some alternatives and is not yet publicly available.
Google’s Gemini 4 Carbon, running in its high-reasoning setting, ranks first on Artificial Analysis’s Intelligence Index v5 with 59 points, one ahead of Claude Opus 5.5 and two ahead of GPT-6 Astra. The result places Carbon at the top of a nine-test benchmark, but does not make it the leader on every task—and Google has not yet opened the model to the public.
The index’s reported scores are 59 for Gemini 4 Carbon, 58 for Claude Opus 5.5 in max mode, 57 for GPT-6 Astra in max mode and 55 for GPT-6.1 Sol in max mode. Google’s earlier Gemini 4 Argon scores 50. Artificial Analysis gives each of four benchmark categories—knowledge work, coding, science and general ability—equal weight in the combined score.
Carbon ranks first on four of the nine tests: HLE-Diamond, Terminal-Bench-Science, AA-Omniscience and AutomationBench-AA. It does not lead the other five. Claude models score higher on office-style knowledge work, Claude Opus 5.5 leads the cited real-world coding test, GPT-6 Astra leads on the physics research test, and Meta’s Muse Wonder Full leads on document reasoning. The overall score therefore compresses different strengths and weaknesses into one ranking.
The source reports an estimated $7.76 per index task at standard prices. That is below the stated $10.04 for Claude Opus 5.5 max, but above GPT-6 Astra max at $6.24 and GPT-6.1 Sol max at $1.27. Google’s introductory API price is listed as $2 per million input tokens and $10 per million output tokens, with later rates of $4 and $20. The source does not specify when those introductory prices begin.
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 With Real Trade-Offs
For businesses choosing models, the result is a reminder that the top overall score is not the same as the best model for a particular job. Carbon’s strengths include science workflows and a test that rewards models for abstaining rather than guessing. A company focused on coding, document work or office tasks may get a better result from a competitor, while a company running long, difficult research tasks may value Carbon’s benchmark performance.
Price comparisons also depend on what a task actually requires. The index’s reported per-task figure gives a useful comparison within this benchmark, but it does not establish what a company’s own work will cost. Task length, tool use, output requirements and the model setting can affect real-world bills. Buyers should compare cost per successfully completed task, not just API rates or leaderboard positions.
For everyday users, the immediate effect is limited: Carbon is not publicly available. If it reaches consumer products later, its reported advantages are most relevant to demanding, multi-step tasks rather than a claim that routine chat will improve across the board. No release date or consumer access terms are provided in the source.
Products Worth Considering
Samsung 27-Inch Odyssey QD-OLED G8 4K 240Hz Gaming Monitor
Samsung Galaxy Tab S10 FE 128GB Wi-Fi Android Tablet, Blue
Sony WH-1000XM6 Wireless Noise-Canceling Over-Ear Headphones, Blue, Bundle
As an affiliate, we earn on qualifying purchases.
How Artificial Analysis Built Version 5
Artificial Analysis launched Index v5 on October 27, replacing its earlier leaderboard structure with nine tests grouped into four equally weighted categories. Knowledge work includes AA-Briefcase, AutomationBench-AA and GDP.pdf; coding includes FrontierCode 2.0 and Terminal-Bench 4.0; science includes Terminal-Bench-Science and the hard subset of CritPt v2; general ability includes AA-Omniscience and HLE-Diamond.
The source says 60% of the index uses private test sets or held-out answers, intended to reduce the chance that models have memorized questions. Many evaluations involve extended agent work, such as operating in a terminal, producing reports or modifying code. That makes the benchmark more task-oriented than a simple set of short questions, though results still reflect only the tests, model versions and settings used.
Carbon’s win on AA-Omniscience needs particular care in interpretation. The source reports 58% accuracy for Carbon, below several models, including Claude Fable 5.1 at 68%, Claude Opus 5.5 at 66%, and GPT-6.1 Sol and GPT-6 Astra at 62%. Carbon nevertheless leads the test’s score because wrong answers are penalized and it has a reported 11% hallucination rate, compared with about 50% for Claude Opus 5.5 and GPT-6 Astra. The measure is not plain factual accuracy: it rewards caution as well as correct answers.
What the Score Leaves Open
The source describes the index and its scores but does not include a direct statement from Google or a complete account of Artificial Analysis’s response to the results. The supplied material also ends mid-sentence during a quotation about coding-agent performance, so that claim and its surrounding context cannot be verified from the text provided.
Several practical details remain unknown. Google has not given a date for broader access, explained how long each rollout stage will last or specified which defenders and partners qualify. The source says Carbon thinks slowly but provides no latency measurements or comparison window. It also gives no underlying methodology for the estimated per-task costs beyond the stated prices and benchmark setting.
The ranking is a result on nine selected tests, not a universal measure of intelligence or a prediction that Carbon will outperform competitors in every deployment. The source does not supply confidence intervals, repeated-run variation or enough detail here to establish how sensitive the one-point gap is to test variability. That makes the small margin a reason to read the component scores, not just the headline position.
Access and Independent Testing
According to the source, Google plans to give Carbon first to a subset of cyber defenders, then to trusted partners, followed by developers, businesses and consumers. No dates are attached to those stages. The next concrete milestones are broader availability details, published pricing in effect at launch and further evaluations that show how the model performs outside the nine index tests.
Companies considering Carbon can test it against their own workloads once access is offered, comparing success rates, time and total cost with alternatives. Readers should also watch for the complete coding-agent comparison and the missing details behind the latency and per-task cost claims. Until those points are clearer and public access begins, the ranking is a benchmark result—not a change users can act on today.
Key Questions
What rank did Gemini 4 Carbon receive?
It scored 59 points on Artificial Analysis’s Intelligence Index v5, ahead of Claude Opus 5.5 at 58 and GPT-6 Astra at 57.
Does Carbon lead all nine tests?
No. The source says Carbon ranks first on four of nine tests and trails rivals on five, including office-style knowledge work and real-world coding.
Why can Carbon lead the Omniscience test with lower accuracy?
The test penalizes wrong answers. Carbon’s reported accuracy is 58%, but its reported hallucination rate is 11%, so its cautious answers raise its score under that test’s scoring approach.
Can the public use Gemini 4 Carbon now?
No public release is reported. Google is said to be starting with a subset of cyber defenders and then expanding access, but no timetable is given.
How much does Carbon cost?
The source lists an estimated $7.76 per index task at standard prices and introductory API rates of $2 per million input tokens and $10 per million output tokens. It does not state when introductory pricing begins.
Source: ThorstenMeyerAI.com
