🔍 Read the full analysis: Gemini 4 Carbon’s AI Score Leads By One Point—Here’s The Broader Picture on ThorstenMeyerAI.com
Get the latest gadgets 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 leads Artificial Analysis’s Intelligence Index v5 with 59 points, one ahead of Claude Opus 5.5 and two ahead of GPT-6 Astra. The ranking reflects performance across nine tests, not a universal measure of AI ability; Carbon leads four tests, has trade-offs in cost and other tasks, and is not yet publicly available.
Google’s Gemini 4 Carbon leads the Artificial Analysis Intelligence Index v5 with 59 points, one point ahead of Claude Opus 5.5 and two ahead of GPT-6 Astra, according to the benchmark results reported October 29. The ranking puts Carbon first on this nine-test index, but the model is not yet publicly available and its narrow lead does not mean it outperforms rivals on every kind of work.
Artificial Analysis launched the rebuilt index on October 27. The supplied results list Gemini 4 Carbon, using its high-reasoning setting, at 59 points; Claude Opus 5.5 at 58; GPT-6 Astra at 57; GPT-6.1 Sol at 55; and Google’s previous Gemini 4 Argon at 50. The index combines nine tests, grouped into knowledge work, coding, science and general capabilities, with each group accounting for 25% of the total.
Carbon places first in four tests: HLE-Diamond, a hard reasoning and knowledge exam; Terminal-Bench-Science, which evaluates science workflows in a terminal; AA-Omniscience, which penalizes incorrect answers; and AutomationBench-AA, which tests business workflows using software tools. Its other results are mixed: the source says rival models lead on office-style knowledge work, real-world coding, physics research, terminal tasks and document reasoning.
The source puts Carbon’s estimated cost at $7.76 per index task at standard prices. That is below the cited $10.04 for Claude Opus 5.5, but above GPT-6 Astra’s $6.24 and GPT-6.1 Sol’s $1.27. Google’s stated introductory API price is $2 per million input tokens and $10 per million output tokens, later rising to $4 and $20, respectively. Those are separate price measures from the index’s estimated cost per task.
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
The result matters because the index offers a side-by-side snapshot of demanding tasks, including agent workflows that can involve using software, working in a terminal or handling extended assignments. It may help organizations identify models worth testing for particular jobs. But the one-point margin between Carbon and Claude Opus 5.5 is small, and the reported task-level results show that the best choice changes with the work.
Cost also complicates a simple ranking-based decision. A model that scores well overall may not be the least expensive option for a company’s actual tasks, while a lower-ranked model may be more capable or economical on a specific workflow. The source’s practical implication is to compare cost per completed task and performance on a company’s own work, rather than switching providers based on the overall score alone.
For individual users, there is no immediate product change: Carbon is not open to the public yet. If it reaches general availability, the reported results suggest its relative strengths may show up most on difficult, long-running tasks, rather than ordinary chat. That is an interpretation of the benchmark, not a guarantee about a future consumer product.
Products Worth Considering
Samsung 27 Odyssey OLED G5 (G50SF) QHD QD-OLED Gaming Monitor
Amazon Kindle Scribe Colorsoft 64GB (1st Generation)
LG 34GX90SA-W 34-inch Ultragear WQHD OLED Curved Gaming Monitor
As an affiliate, we earn on qualifying purchases.
How Artificial Analysis Scores Models
The v5 index uses nine tests split evenly among four categories. Knowledge work accounts for 25% and includes AA-Briefcase, AutomationBench-AA and GDP.pdf. Coding contributes 25% through FrontierCode 2.0 and Terminal-Bench 4.0. Science contributes another 25% through Terminal-Bench-Science and the hard subset of CritPt v2. The final quarter covers general performance through AA-Omniscience v1.1 and HLE-Diamond.
Artificial Analysis says 60% of v5 uses private test sets or held-out answers, and many tasks involve longer work such as producing reports, operating in a terminal or fixing code. That design aims to test performance beyond short question-and-answer exchanges. The index remains a particular test suite: its overall score represents results on these tasks and settings, not a definitive ranking of every model for every use.
One result illustrates how test design affects a ranking. The source reports that Carbon answered 58% of the questions correctly on AA-Omniscience, below several Claude and GPT models, but scored highest on the test because it made fewer wrong guesses. Artificial Analysis reports an 11% hallucination rate for Carbon under that test’s measure, compared with about 50% for Claude Opus 5.5 and GPT-6 Astra. The score rewards cautious answers; it does not show that Carbon has the broadest knowledge.
“choose by task and by cost per finished task, not by rank.”
— Thorsten Meyer, source article
Access and Real-World Performance
Google has not given a public release date for Gemini 4 Carbon. The source describes a staged rollout, starting with a subset of cyber defenders, followed by trusted partners and then developers, businesses and consumers. The timing and conditions for each stage remain unspecified.
The supplied material also does not provide the full underlying benchmark tables or explain every evaluation setting, so the reported figures should be read as attributed results rather than independently verified here. It does not establish how Carbon will perform on any particular company’s data, how its API prices will apply to a specific workload, or whether the one-point index margin is meaningful beyond the stated scoring system. Benchmark rankings can also change when models, test sets or evaluation settings change.
Google’s Staged Rollout
According to the source, Google plans to provide Carbon first to a subset of cyber defenders, then to trusted partners, with developers, businesses and consumers to follow. No dates or access criteria for those later stages were provided. The next practical developments to watch are Google’s access announcements and the release of more information about the model’s availability and API pricing.
For organizations, the near-term step is to wait for access and test the model against their own tasks, accuracy requirements and costs before making a purchasing decision. For consumers, the index result alone does not signal an immediate change to Google’s publicly available products.
Key Questions
What did Gemini 4 Carbon score?
It scored 59 points on Artificial Analysis’s Intelligence Index v5, one point ahead of Claude Opus 5.5 and two ahead of GPT-6 Astra, according to the reported results.
Does the ranking mean Carbon is better at every task?
No. The index combines nine tests, and the source says Carbon leads four while rival models lead in areas including office-style knowledge work and real-world coding. The result is specific to the index’s tests and scoring.
Can the public use Gemini 4 Carbon now?
No public access date has been announced. The source says Google is starting with a subset of cyber defenders, followed by trusted partners and later wider groups.
How much does Carbon cost?
The source estimates $7.76 per index task at standard prices. It also reports an introductory API price of $2 per million input tokens and $10 per million output tokens, later rising to $4 and $20. The source does not specify a date for that price change.
Why did Carbon lead the AA-Omniscience test?
The source says Carbon did not have the highest plain-answer accuracy, but it made fewer incorrect guesses. Because AA-Omniscience penalizes wrong answers, Carbon’s reported lower hallucination rate helped it score highest on that test.
Source: ThorstenMeyerAI.com
