🔍 Read the full analysis: Gemini 4 Carbon Takes The New #1 AI Model Spot By One Point—What Rankings Leave Out on ThorstenMeyerAI.com
Get movie nights 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’ 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, has a higher stated task cost than some competitors, and is not yet publicly available.
Google’s Gemini 4 Carbon has taken the top spot on Artificial Analysis’ Intelligence Index v5, scoring 59 points—one more than Claude Opus 5.5 and two more than GPT-6 Astra. The ranking is a narrow lead across nine selected tests, not a general verdict on which AI model is best; Carbon is also not yet publicly available, with Google starting access among a limited group of cyber defenders.
Artificial Analysis launched version 5 of its index on October 27, then listed Carbon at the top in results reported the following day. The scores cited by ThorstenMeyerAI.com are for Carbon at its high-reasoning setting, Claude Opus 5.5 at max, and GPT-6 Astra at max. Claude scored 58, GPT-6 Astra 57, GPT-6.1 Sol 55, and Google’s previous model, Gemini 4 Argon, 50.
Carbon placed first in four of the nine tests, including hard reasoning and knowledge questions, science workflows performed in a terminal, a knowledge test that penalizes wrong answers, and business workflows across software tools. It did not lead the other five. The source reports that Claude models scored higher on office-style knowledge work, Claude Opus 5.5 led on real-world coding, GPT-6 Astra led on a physics research test, Kimi K3.1 Pro led on terminal tasks, and Meta’s Muse Wonder Full led on document reasoning.
The index assigns equal weight to four categories: knowledge work, coding, science and general performance. The source says the reported cost per index task is $7.76 for Carbon, compared with $10.04 for Claude Opus 5.5 and $6.24 for GPT-6 Astra. These figures are based on the source’s stated standard prices and benchmark workload; they do not establish what a particular business would pay for its own tasks.
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 Has Limits
The result gives Google a strong position in a closely watched model comparison, but the one-point margin is small and the score combines performance across a particular set of tests. A model that ranks first overall can still perform worse on work that matters to a given user or company. The leaderboard itself therefore does not settle which model should be used for coding, document analysis, research or routine office tasks.
Cost and availability also affect the practical value of the ranking. On the figures reported by ThorstenMeyerAI.com, Carbon costs less per index task than Claude Opus 5.5 but more than GPT-6 Astra and substantially more than GPT-6.1 Sol. Those benchmark-level prices are a point of comparison, not a substitute for measuring the cost of completing a company’s own work. Teams weighing a switch would need to test their workflows, quality requirements and costs directly.
For everyday users, there is no immediate change: Carbon has not been opened to the public. If Google’s staged rollout proceeds, the model’s performance on longer, more demanding tasks may matter more than its narrow lead on a single composite score.
Popular right now
De'Longhi Magnifica Evo Automatic Espresso Machine with Milk Frother – Silver
Ninja Flip Toaster Oven & Air Fryer, 8-in-1, Stainless Steel, SP151
Ninja Foodi 10-Quart DualZone 6-in-1 Air Fryer, Black, DZ302
As an affiliate, we earn on qualifying purchases.
How the New Index Scores Models
Artificial Analysis describes the Intelligence Index as a combined score drawn from nine tests, rather than a direct measure of general intelligence. Version 5 divides those tests among four equally weighted blocks. Within the blocks, the individual tests have different weights; for example, the source lists FrontierCode 2.0 at 15% of the total and Terminal-Bench at 10%.
The tests include work such as producing reports, completing software workflows and carrying out terminal-based coding or science tasks. ThorstenMeyerAI.com says 60% of the index uses private test sets or held-out answers, intended to reduce the effect of models having memorized questions. The index measures models in particular configurations on these tasks, so changing the test mix or model settings could change the order.
One of Carbon’s reported wins illustrates how test design matters. On AA-Omniscience, Carbon scored 54, six points ahead of Claude Opus 5.5, despite having lower plain-answer accuracy than several models. The source says the test penalizes incorrect answers, and reports an 11% hallucination rate for Carbon, against about 50% for Claude Opus 5.5 and GPT-6 Astra. The score thus reflects both answers given and the benchmark’s treatment of wrong answers; it should not be read as proof that Carbon knows more.
“a subset of cyber defenders”
— Google, as reported by ThorstenMeyerAI.com
Access and Real-World Results
The source says Google plans to provide Carbon first to a limited group of cyber defenders, then to trusted partners, developers, businesses and consumers. No dates are given for those stages, and it is not clear when general users will be able to try the model. The source also reports introductory API pricing of $2 per million input tokens and $10 per million output tokens, rising later to $4 and $20 respectively, but does not specify when that change would take effect.
The leaderboard does not show how Carbon will perform across every real-world workload, nor whether the one-point gap from Claude Opus 5.5 would persist under other settings or evaluation methods. The supplied source text also ends partway through a comparison of coding-agent results, so its full account of that comparison cannot be verified here. Benchmark scores and cost-per-task figures should be treated as reported results, not as a guarantee of performance or price for an individual deployment.
Google’s Staged Release Plans
The next reported milestone is Google’s rollout to a subset of cyber defenders, followed by trusted partners, developers, businesses and consumers. The source does not provide a calendar for those steps. Until access broadens, public users and most businesses cannot independently test Carbon against the tasks they care about.
For organizations evaluating the result, the practical next step is to compare candidate models on their own workloads, including the quality of completed work and its total cost. Further leaderboard updates, release details and independent testing could clarify whether Carbon’s lead holds beyond the nine tests in version 5. For consumers, the key change to watch for is public availability; no immediate product access is reported.
Key Questions
What score did Gemini 4 Carbon receive?
It scored 59 points on Artificial Analysis’ Intelligence Index v5, one point ahead of Claude Opus 5.5 and two ahead of GPT-6 Astra, according to the source.
Does the ranking mean Carbon is the best AI model for every task?
No. The index combines results from nine tests. Carbon led four of them, while other models led on office-style work, coding, physics research, terminal tasks or document reasoning.
Can the public use Gemini 4 Carbon now?
Not according to the source. Google is reported to be starting with a subset of cyber defenders, with later access planned for other groups but no rollout dates provided.
How does Carbon’s reported benchmark cost compare with rivals?
The source gives a cost of $7.76 per index task for Carbon, versus $10.04 for Claude Opus 5.5 and $6.24 for GPT-6 Astra. Those are reported benchmark-cost comparisons and may not match the cost of a user’s specific workload.
Source: ThorstenMeyerAI.com
