🔍 Read the full analysis: Gemini 4 Carbon’s One-Point Lead: Reading The AI Model Ranking Carefully on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- 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 placed first in four of nine tests, but lost five, costs more per index task than some rivals and is not yet publicly available.
Google’s Gemini 4 Carbon has taken the top position in Artificial Analysis’s Intelligence Index v5, scoring 59 points—one ahead of Claude Opus 5.5 and two ahead of GPT-6 Astra. The narrow lead marks a ranking result, not a general verdict on which AI model is best: Carbon placed first on four of the index’s nine tests and did not lead the other five.
Artificial Analysis released version 5 of its index on October 27, then updated the leaderboard the following day with Carbon at the top. The reported scores are 59 for Carbon, 58 for Claude Opus 5.5 at maximum reasoning, 57 for GPT-6 Astra at maximum reasoning and 55 for GPT-6.1 Sol at maximum reasoning. Google’s previous model, Gemini 4 Argon, scored 50.
Carbon led four individual tests: HLE-Diamond, for difficult reasoning and knowledge questions; Terminal-Bench-Science, for science work in a terminal; AA-Omniscience, which penalizes wrong answers; and AutomationBench-AA, which tests business workflows. Its results included 69% on HLE-Diamond, compared with 68% for Claude Opus 5.5, and 65% on Terminal-Bench-Science, compared with 63% for GPT-6 Astra.
The model fell behind competitors on the other five tests. Claude models led office-style knowledge work, Claude Opus 5.5 led real-world coding, GPT-6 Astra led a physics research test, Kimi K3.1 Pro led terminal tasks and Meta’s Muse Wonder Full led document reasoning. Artificial Analysis’s figures put Carbon’s cost at $7.76 per index task at standard prices, compared with $10.04 for Claude Opus 5.5, $6.24 for GPT-6 Astra and $1.27 for GPT-6.1 Sol.
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 Trade-Offs
The result matters because a one-point margin can shape how buyers and developers compare leading models, but the index’s overall score compresses different abilities into a single ranking. Carbon’s position does not mean it leads at every kind of work. Its test results point instead to a specific mix of strengths, including science workflows, business automation and cautious answering on a test that penalizes incorrect responses.
Cost also changes the practical comparison. Carbon’s stated index-task price is lower than Claude Opus 5.5’s but higher than GPT-6 Astra’s and substantially higher than GPT-6.1 Sol’s. These are costs for the benchmark’s workload, not a universal price for every customer task. Companies would need to compare the cost per successful task on their own work, including any differences in accuracy, speed and tool use, before treating the leaderboard as a purchasing guide.
For consumers, there is no immediate product change: Carbon is not publicly available. Google’s initial rollout is limited to a subset of cyber defenders, with access planned for trusted partners and later for developers, businesses and consumers. The company has not supplied a public date for those later stages.
Products Worth Considering
LG UltraGear 32GX850A-B 32-Inch 4K UHD Glossy OLED Gaming Monitor
DJI Osmo Pocket 3 Vlogging Camera with 1-Inch CMOS Sensor and 4K Video
Sony WH-1000XM6 Wireless Noise-Canceling Over-Ear Headphones, Blue, Bundle
As an affiliate, we earn on qualifying purchases.
How Version Five Scores Models
Artificial Analysis describes the Intelligence Index as a combined score across nine tests, organized into four equally weighted blocks: knowledge work, coding, science and general capability. The individual tests have different weights within those blocks. The index is intended to compare model performance on its chosen tasks and settings; it is not a direct measurement of intelligence across every possible use.
According to the source material, 60% of version five uses private test sets or held-out answers. Many assessments involve extended agent work, such as operating in a terminal, producing reports or modifying code, rather than answering only short prompts. That can make the results more representative of certain work tasks, while also making benchmark runs more expensive.
One of Carbon’s wins illustrates how test design can affect rankings. On AA-Omniscience, Carbon scored 54 points, six ahead of Claude Opus 5.5. Yet the source reports that Carbon answered 58% correctly, below Claude Fable 5.1 at 68%, Claude Opus 5.5 at 66%, and GPT-6.1 Sol and GPT-6 Astra at 62% each. The test also penalizes wrong answers: Artificial Analysis reports an 11% hallucination rate for Carbon, versus about 50% for Claude Opus 5.5 and GPT-6 Astra under its definition. Carbon’s higher score on that test reflects its lower rate of wrong answers, not the highest raw accuracy.
“Choose by task and by cost per finished task, not by rank.”
— Thorsten Meyer, in the supplied source article
What the Leaderboard Cannot Settle
The supplied results establish Carbon’s position on this version of this index, but do not show how it performs across every business, consumer or technical workload. The overall lead is one point, and the order depends on the tests, weights and model settings used. The source does not provide uncertainty ranges or enough information here to determine whether that margin is larger than normal run-to-run variation.
Some comparisons also involve different model settings, including maximum reasoning for the named Claude and GPT systems. The supplied material does not detail all settings, prompt configurations or repeat-run procedures. Nor does it provide a full speed comparison, though it describes Carbon as slow to think. That makes it difficult to judge the combined effect of latency, price and performance for a particular use case from the headline score alone.
Availability and pricing may change. The source gives introductory API pricing of $2 per million input tokens and $10 per million output tokens, later rising to $4 and $20, respectively, but does not state when those prices take effect or when public access will begin. Those figures should not be confused with the benchmark’s reported cost per index task.
Rollout and Real-World Testing
Google’s stated sequence begins with access for a subset of cyber defenders, followed by trusted partners, then developers, businesses and consumers. No dates are provided for those stages. Until access expands, independent customers cannot directly evaluate Carbon against their own systems and tasks.
The next useful evidence will come from broader availability and task-specific testing: how Carbon performs on organizations’ actual workflows, how often it completes them correctly, how long it takes and what each successful result costs. Artificial Analysis may also update its leaderboard as models or benchmark results change. For now, the reported finding is precise but limited: Carbon leads this index by one point, while the wider choice between models depends on the job and the price of completing it.
Key Questions
What is Gemini 4 Carbon’s score?
Carbon 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.
Does Carbon lead every test in the index?
No. The source says Carbon placed first on four of nine tests and lost the other five, including tests for office-style knowledge work and real-world coding.
Can the public use Gemini 4 Carbon now?
No. Google is initially giving access to a subset of cyber defenders, then plans to expand to partners, developers, businesses and consumers. No public release date is given.
Is Carbon cheaper than its closest competitors?
It depends on the comparison. The supplied figures list $7.76 per index task for Carbon, less than Claude Opus 5.5 at $10.04 but more than GPT-6 Astra at $6.24 and GPT-6.1 Sol at $1.27. Benchmark-task costs are not a universal estimate for every customer workload.
Does the ranking prove Carbon is the best AI model?
No. It shows that Carbon had the highest combined score on Artificial Analysis’s nine selected tests at the reported settings. Results may differ on other tasks, tests or settings.
Source: ThorstenMeyerAI.com
