🔍 Read the full analysis: Gemini 4 Carbon Is #1 By One Point: A Closer Look At The AI Model Score on ThorstenMeyerAI.com
Prime made for students and young adults
- Fast, free delivery for dorm and study essentials
- Prime Video and Amazon Music included
- Member-only deals
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 the nine tests, but is not the strongest across every category, and Google has not opened it to the public.
Google’s Gemini 4 Carbon scored 59 points to take first place on the Artificial Analysis Intelligence Index v5, one point ahead of Claude Opus 5.5 and two ahead of GPT-6 Astra. The ranking is a narrow result across nine tests, not a general verdict that Carbon is the strongest model for every task; Google says access is currently limited to a subset of cyber defenders.
Artificial Analysis rebuilt its leaderboard on October 27, introducing version 5 of the index. The scores listed in the source are 59 for Gemini 4 Carbon, 58 for Claude Opus 5.5, 57 for GPT-6 Astra, 55 for GPT-6.1 Sol and 50 for Google’s previous model, Gemini 4 Argon. Carbon’s score is for its high-reasoning setting; the rival scores are for the settings specified in the ranking.
Carbon placed first on four of the nine component tests: HLE-Diamond, Terminal-Bench-Science, AA-Omniscience and AutomationBench-AA. It did not lead the other five. The source reports rival advantages in office-style knowledge work, real-world coding, physics research, terminal tasks and document reasoning. Those results mean the index’s overall order compresses meaningful differences between task types into a single score.
The source puts Carbon’s cost at $7.76 per index task at standard prices. That is below the stated $10.04 for Claude Opus 5.5, but above GPT-6 Astra at $6.24 and GPT-6.1 Sol at $1.27. Google’s stated introductory API pricing is $2 per million input tokens and $10 per million output tokens, later rising to $4 and $20 respectively. The source does not give a public release date.
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 a model, the result is a reason to compare performance on their own work rather than treating a leaderboard position as a purchasing recommendation. Carbon’s one-point lead reflects a particular mix of benchmarks and model settings. A company focused on coding or office documents could find a different model better suited to its workload, while the cost of completing a task may matter more than the overall score.
The ranking also highlights how answering cautiously can affect benchmark results. On AA-Omniscience, Carbon scored 54 points, six ahead of Claude Opus 5.5. Yet the source says Carbon’s plain accuracy was 58%, below several models, including Claude Fable 5.1 at 68% and Claude Opus 5.5 at 66%. The test penalizes incorrect answers, and the reported hallucination rate—defined by Artificial Analysis as wrong answers among questions not answered correctly—was 11% for Carbon versus about 50% for Claude Opus 5.5 and GPT-6 Astra. That measures performance under this test’s scoring rules, not every dimension of factual reliability.
For ordinary users, the immediate effect is limited: Carbon is not publicly available according to the source. If Google broadens access, its stronger results on some long, demanding tasks could matter, but the index alone does not establish how it will perform in everyday consumer use.
Popular right now
Bose QuietComfort Wireless Noise-Cancelling Over-Ear Headphones, Black
Dyson Car+Boat Cordless Handheld Vacuum
KitchenAid Artisan 5-Quart Tilt-Head Stand Mixer, Almond Cream
As an affiliate, we earn on qualifying purchases.
How Version Five Scores Models
Artificial Analysis describes its Intelligence Index as a combined score from nine tests grouped into four equally weighted blocks: knowledge work, coding, science and general capability. The individual tests have different weights within those blocks. The index includes tasks such as generating reports, using software tools, completing terminal-based work and answering difficult knowledge questions.
According to the source, 60% of version 5 uses private test sets or held-out answers, intended to reduce the effect of models having memorized questions. Many tasks are long-running agent workflows, which the source says are more realistic than short prompts but also more expensive to run. The total score still reflects only these selected tests, their weighting and the particular model settings used.
The timing is unusually close: the source says Artificial Analysis rebuilt the leaderboard on Tuesday, October 27, and Carbon took the top spot the following day. Gemini 4 Argon scored 50 in the v5 comparison, below Carbon’s 59, but that difference is a comparison within this index rather than evidence of a uniform improvement across all uses.
Access and Benchmark Limits
Public availability remains undated. The source describes a staged plan—cyber defenders first, then trusted partners, developers, businesses and consumers—but supplies no schedule for those steps. It is also not clear from the material how many testers are included in the initial group or what access conditions apply.
The supplied account does not include the full results for every test or independent confirmation of the benchmark figures. It also does not establish how the score translates to particular products, customer workloads or real-world reliability. The stated task costs and API prices are reported figures; actual spending will depend on usage and task requirements.
Google’s Planned Access Rollout
Google’s stated next step is to provide Carbon to a subset of cyber defenders, followed by trusted partners and then wider groups, including developers, businesses and consumers. No dates for those stages are given in the source, so when the public will be able to try the model remains unknown.
For now, companies considering a switch can compare models on representative internal tasks and account for cost per completed task, not just the leaderboard score. The next useful developments will be details about wider access, independent use outside the benchmark and performance across specific applications.
Key Questions
What score did Gemini 4 Carbon receive?
Carbon scored 59 points on the Artificial Analysis Intelligence Index v5, one point ahead of Claude Opus 5.5 and two ahead of GPT-6 Astra.
Does the index show that Carbon is best at every task?
No. Carbon ranked first on four of nine tests, while other models led in areas including office-style knowledge work and real-world coding.
Can the public use Gemini 4 Carbon now?
Not according to the source material. Google is starting with a subset of cyber defenders and has not announced a date for public access.
How much does Carbon cost to run?
The source reports an average of $7.76 per index task at standard prices. It also lists introductory API rates of $2 per million input tokens and $10 per million output tokens, with later rates of $4 and $20.
Why did Carbon lead the AA-Omniscience test?
The test penalizes wrong answers. The source says Carbon had lower plain accuracy than several rivals but a lower reported hallucination rate, which helped it score higher under that test’s scoring method.
Source: ThorstenMeyerAI.com
