🔍 Read the full analysis: Gemini 4 Carbon’s One-Point Edge Changes The AI Model Rankings 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’ Intelligence Index v5, one point ahead of Claude Opus 5.5 and two ahead of GPT-6 Astra. The ranking reflects nine tests, not a universal measure of capability: Carbon leads four tests, loses five, and is not yet publicly available.
Google’s Gemini 4 Carbon has taken the top spot in Artificial Analysis’ Intelligence Index v5, scoring 59 points, one more than Claude Opus 5.5 and two more than GPT-6 Astra. The result puts Carbon first on this particular nine-test benchmark, but the model is not yet publicly available: Google is initially releasing it to a subset of cyber defenders, according to the supplied report.
The ranking follows Artificial Analysis’ October 27 launch of version 5 of its index. In the reported scores, Claude Opus 5.5 (maximum reasoning setting) earned 58, GPT-6 Astra (maximum) scored 57, and GPT-6.1 Sol (maximum) reached 55. Google’s earlier Gemini 4 Argon scored 50. These results compare models at specified settings; they do not establish that one model is best for every use.
Carbon placed first in four of the nine tests: HLE-Diamond, Terminal-Bench-Science, AA-Omniscience and AutomationBench-AA. It did not lead the other five. The reported leaders elsewhere include Claude models for office-style work and real-world coding, GPT-6 Astra for physics research, Kimi K3.1 Pro for terminal tasks, and Meta’s Muse Wonder Full for document reasoning.
The reported cost for Carbon was $7.76 per index task at standard prices. That is below Claude Opus 5.5’s $10.04, but above GPT-6 Astra’s $6.24 and GPT-6.1 Sol’s $1.27. The source also lists an introductory API price of $2 per million input tokens and $10 per million output tokens, rising later to $4 and $20 respectively. Those prices and the task-cost estimate describe different measures and should not be treated as interchangeable.
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
A one-point advantage can affect which model appears first on a leaderboard, but it is a small margin in a composite score. For organizations selecting an AI system, the more practical question is whether a model performs well on their actual work and what it costs to complete a task reliably. The reported spread in task results shows why the overall ranking alone is not a procurement decision.
Carbon’s reported result on AA-Omniscience also illustrates how a benchmark’s scoring rules shape the outcome. It scored 54 on that test, six points ahead of Claude Opus 5.5, despite lower plain-answer accuracy in the figures supplied. Artificial Analysis’ metric penalizes wrong answers, and the report attributes Carbon’s advantage partly to its lower stated hallucination rate: 11%, compared with about 50% for Claude Opus 5.5 and GPT-6 Astra. That favors caution on this test; it does not mean Carbon knows more overall or will be more accurate in every setting.
For everyday users, the immediate effect is limited because access is restricted. For businesses and developers, the score may make Carbon worth testing when it becomes available, particularly for tasks represented in its stronger benchmark results. The reported cost differences also make cost per successful task, rather than headline rank or token price alone, a relevant comparison.
Products Worth Considering
Bose QuietComfort True Wireless Earbuds with Active Noise Cancellation, Black
Apple iPad Air (4th Generation, 10.9-inch, Wi-Fi, 256GB) – Space Gray
DJI Osmo Pocket 3 Vlogging Camera with 1-Inch CMOS Sensor and 4K Video
As an affiliate, we earn on qualifying purchases.
How Version Five Scores Models
Artificial Analysis describes its index as a combined score from nine tests arranged in four equally weighted blocks: knowledge work, coding, science and general capability. The supplied breakdown gives 25% weight to each block, with individual tests carrying different shares within them. For example, FrontierCode 2.0 accounts for 15% of the overall score, while AutomationBench-AA accounts for 5%.
The tests include activities such as producing reports, working with software tools, conducting terminal-based science workflows and solving difficult reasoning questions. According to the report, 60% of version 5 uses private test sets or held-out answers, intended to reduce the effect of models having seen the questions during training. The index is therefore a measure of performance on its selected tasks and settings—not a complete assessment of intelligence, safety, usability or performance across all real-world work.
“a subset of cyber defenders”
— Google, as quoted in the supplied report
Access Dates and Real-World Performance
The supplied information gives no date for public availability. Google’s reported rollout sequence is cyber defenders, trusted partners, then developers, businesses and consumers, but timing for those stages is unspecified. The report does not provide independent confirmation of Google’s rollout schedule or explain how many users are included in the initial group.
The benchmark figures also do not settle how Carbon will perform across a particular company’s workload, or how the reported costs will translate into its own usage. The source material’s account of Carbon’s losses is incomplete, and does not provide every underlying score for all nine tests. Rankings can change with model settings, benchmark design and task selection, so the one-point overall margin should be read within those limits.
Google’s Staged Release Ahead
According to the supplied report, Google plans to expand Carbon access from a subset of cyber defenders to trusted partners, then to developers, businesses and consumers. No dates were given for those stages. The next concrete development to watch is whether Google announces wider access, publishes fuller product and pricing details, or provides additional information about the model’s performance and availability.
Until then, prospective users can treat the index as a reason to evaluate Carbon, not as a substitute for testing. Comparisons on their own tasks—measuring output quality, error rates, speed and total cost—will show whether its benchmark strengths apply to their needs. For consumers, the ranking does not change what is available today.
Key Questions
What rank did Gemini 4 Carbon receive?
It scored 59 points on Artificial Analysis’ Intelligence Index v5, ahead of Claude Opus 5.5 at 58 and GPT-6 Astra at 57.
Does the result mean Carbon is the best AI model for every task?
No. The ranking combines nine tests. The supplied report says Carbon led four and lost five, with different models leading in office work, coding, physics research, terminal tasks and document reasoning.
Can the public use Gemini 4 Carbon now?
Not according to the supplied report. Google is starting with a subset of cyber defenders and has not announced dates for broader access.
How much does Carbon cost?
The report lists an introductory API price of $2 per million input tokens and $10 per million output tokens, later rising to $4 and $20. It separately estimates $7.76 per index task at standard prices.
Why did Carbon lead the AA-Omniscience test?
The supplied report says the test penalizes incorrect answers. Carbon had lower plain-answer accuracy than several listed competitors but a reported 11% hallucination rate, which helped it score higher under that test’s penalty system.
Source: ThorstenMeyerAI.com
