🔍 Read the full analysis: Beyond #1: What Gemini 4 Carbon’s One-Point AI Ranking Leaves Unanswered 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 above Claude Opus 5.5 and two above GPT-6 Astra. The ranking reflects nine selected tests, not a universal measure of AI ability; Carbon is not publicly available yet, and its relative cost and performance vary by task.
Google’s Gemini 4 Carbon has taken the top spot on Artificial Analysis’s Intelligence Index v5, scoring 59 points—one more than Claude Opus 5.5 and two more than GPT-6 Astra. The narrow lead marks Carbon as the highest-scoring model on this particular nine-test benchmark, but it does not establish that it is best for every kind of work, and Google has not yet made it publicly available.
The leaderboard followed Artificial Analysis’s October 27 rebuild of its index. In the reported comparison, Claude Opus 5.5 scored 58, GPT-6 Astra 57, GPT-6.1 Sol 55 and Google’s previous model, Gemini 4 Argon, 50. Carbon’s result is listed at its high-reasoning setting; the competing Claude and GPT scores cited are at their maximum settings.
Carbon placed first on four of the nine tests, according to the source material: HLE-Diamond, Terminal-Bench-Science, AA-Omniscience and AutomationBench-AA. Its results were mixed across the other tests. It trailed Claude models on office-style knowledge work and real-world coding, GPT-6 Astra on a physics research test, Kimi K3.1 Pro on terminal tasks, and Meta’s Muse Wonder Full on document reasoning.
The index’s reported estimate is $7.76 per task at standard prices. That is below the cited $10.04 for Claude Opus 5.5 at maximum settings, but above GPT-6 Astra’s $6.24 and well above GPT-6.1 Sol’s $1.27. The source also gives introductory API rates of $2 per million input tokens and $10 per million output tokens, later rising to $4 and $20, respectively. These are reported pricing figures, not a guarantee of what a particular customer’s workload will cost.
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 Winner
A one-point gap gives Carbon the top position on the index, but offers little separation from its closest rivals. The result may matter to organizations comparing advanced models for complex tasks, yet it is not a substitute for testing the specific work they need done. A model that leads on a combined score can still trail on a heavily weighted task that matters to a company.
Cost adds another qualification. Carbon’s cited per-task estimate is lower than Claude Opus 5.5’s, but higher than GPT-6 Astra’s and many times the reported GPT-6.1 Sol figure. For businesses, the relevant comparison is the cost and quality of completing their own tasks, including any errors or extra work—not simply the headline ranking or token price.
For consumers, there is no immediate change: Carbon is not publicly accessible. If it becomes broadly available, the reported results point to possible advantages on demanding reasoning, science and workflow tests, not proof of a better experience in routine chat.
Products Worth Considering
Anker SOLIX S2000 Portable Power Station, 2,010Wh, 1,500W
Apple iPad Air (4th Generation, 10.9-inch, Wi-Fi, 256GB) – Space Gray
Samsung 27-inch Odyssey OLED G6 QD-OLED QHD Gaming Monitor
As an affiliate, we earn on qualifying purchases.
What the New Index Measures
Artificial Analysis describes its Intelligence Index as a combined score from nine tests grouped into four equally weighted areas: knowledge work, coding, science and general performance. The source says the index assigns 25% of the overall score to each area. Individual tests have different weights within those groups, so the result reflects a selected balance of tasks rather than a direct measure of every model capability.
Version 5 launched on October 27. According to the supplied account, 60% of the index uses private test sets or held-out answers, and many tasks involve longer agent-style work such as using a terminal, producing reports or changing code. Those design choices aim to test performance beyond short question-and-answer prompts, while also making runs more involved and expensive.
One result illustrates how test design affects rankings. Carbon scored 58% on plain accuracy in AA-Omniscience, below several Claude and GPT models. It nevertheless led that test’s penalty-adjusted score, with a reported 11% hallucination rate. The source attributes the difference to the test penalizing incorrect answers: a model that abstains more often can score better on that measure than one that answers more questions but gets a larger share wrong. That is a result on this test, not a claim that Carbon is the most accurate model overall.
“choose by task and by cost per finished task, not by rank.”
— Thorsten Meyer, in the supplied source article
Access, Pricing and Real-World Fit
There is no public release date in the source material. Google is described as starting with a subset of cyber defenders, then moving to trusted partners and later to developers, businesses and consumers. The timing and terms of those stages are not specified.
The supplied information also does not provide the full result tables, run-to-run variation, or enough detail to independently assess how the reported per-task cost maps onto different workloads. Benchmark scores may not predict performance on a company’s own data, tools or safety requirements. The one-point overall margin should be read as a close result on this index, not evidence of a meaningful lead in every setting.
Some cited head-to-head details are reported without the full underlying methodology or data in the material provided. Readers therefore cannot determine from these figures alone how stable each test result is, or how much small differences would matter in everyday use.
Google’s Staged Release Plans
The reported rollout is set to begin with a subset of cyber defenders, followed by trusted partners and then wider access for developers, businesses and consumers. Google has not supplied a timetable in the source material, so the next clear milestone is confirmation of availability and release terms.
Organizations considering Carbon can compare it with alternatives on representative tasks, checking answer quality, failure rates, speed and the total cost of completing work. Further Artificial Analysis results or independent evaluations could also clarify how the index standings hold up across repeated testing. Until those details and public access are available, the ranking is a benchmark result—not a reason by itself to switch models.
Key Questions
Did Gemini 4 Carbon rank first overall?
Yes. 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. That is first on this index, not a universal ranking of AI systems.
Is Gemini 4 Carbon available to the public?
Not according to the supplied report. Google is described as beginning with a subset of cyber defenders and planning later access for partners, developers, businesses and consumers, without a public release date.
What does the index score measure?
The index combines results from nine tests across knowledge work, coding, science and general performance. Its score reflects that particular test mix and weighting; it does not cover every task or use case.
Is Carbon cheaper than competing models?
Not across all comparisons. The source estimates $7.76 per index task, below Claude Opus 5.5’s cited $10.04 but above GPT-6 Astra’s $6.24 and GPT-6.1 Sol’s $1.27. Actual costs depend on usage and task.
Source: ThorstenMeyerAI.com
