🔍 Read the full analysis: Gemini 4 Carbon Takes The AI Ranking’s Top Spot—With Just One Point To Spare on ThorstenMeyerAI.com
Get privacy and security gear 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 leads the Artificial Analysis Intelligence Index v5 with 59 points, edging Claude Opus 5.5 by one and GPT-6 Astra by two. The result covers nine tests, not every kind of AI task; Carbon is not publicly available, and its price and performance vary by workload.
Google’s Gemini 4 Carbon has taken the top position on the Artificial Analysis Intelligence Index v5, scoring 59 points—one ahead of Claude Opus 5.5 and two ahead of GPT-6 Astra. The ranking is based on nine tests, and Carbon is not yet publicly available: Google says it is starting with a limited group of cyber defenders.
The result follows a major revision to the index, which Artificial Analysis launched on October 27. Its v5 ranking gives equal weight to four categories—knowledge work, coding, science and general performance—each accounting for 25% of the combined score. Carbon leads the listed models, while Google’s earlier Gemini 4 Argon scores 50. The figures compare specific models at particular settings; they are not a general rating of every version or use of each system.
Carbon places first on four of the nine component tests: HLE-Diamond, Terminal-Bench-Science, AA-Omniscience and AutomationBench-AA. Its reported scores include 69% on HLE-Diamond, against 68% for Claude Opus 5.5, and 65% on Terminal-Bench-Science, against 63% for GPT-6 Astra. It also reaches 78% on AutomationBench-AA, slightly ahead of Gemini 4 Argon.
Its weaknesses are visible in the same results. Carbon trails Claude models on office-style knowledge work and real-world coding, GPT-6 Astra on physics research, Kimi K3.1 Pro on terminal tasks and Meta’s Muse Wonder Full on document reasoning. On FrontierCode 2.0, Carbon scores 41%, compared with 46% for Claude Opus 5.5. The report also lists a cost of $7.76 per index task at standard prices: below Claude Opus 5.5’s $10.04, but above GPT-6 Astra’s $6.24 and GPT-6.1 Sol’s $1.27.
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 Across Mixed Workloads
The one-point margin makes the result a close benchmark contest, not a decisive separation in capability. Carbon’s first-place overall score reflects its combined performance across this particular set of tasks. A buyer whose work resembles the tests where Carbon leads may read the ranking differently from one focused on coding or office-style knowledge work, where other models score higher.
The index’s treatment of uncertainty also affects the result. Artificial Analysis reports that Carbon scores 54 on AA-Omniscience, ahead of Claude Opus 5.5 by six points. Yet Carbon’s plain accuracy on that test is 58%, below several Claude and GPT models. The source says the test penalizes incorrect answers and reports an 11% hallucination rate for Carbon, compared with about 50% for Claude Opus 5.5 and GPT-6 Astra. In this scoring setup, a model that declines to answer when unsure can do better than one that answers more questions but gets more of them wrong.
For organizations, the ranking is a reason to test models on their own work and compare the cost of completed tasks, rather than choosing by the overall position alone. For consumers, the immediate effect is limited: Carbon is not open for general use, and Google has not given a public release date.
Products Worth Considering
Jackery HomePower 3000 Portable Power Station with Two 200W Solar Panels
DJI Osmo Pocket 3 Creator Combo Vlogging Camera with Mic 2, 1 CMOS, 4K/120fps, 3-Axis Stabilization, 128GB Card
Bose QuietComfort Wireless Noise-Cancelling Over-Ear Headphones, Black
As an affiliate, we earn on qualifying purchases.
How Artificial Analysis Built Version 5
Artificial Analysis describes its Intelligence Index as a combined score from nine tests, not a single exam. Version 5 divides the weighting equally among knowledge work, coding, science and general performance. Within those categories, the source lists tests such as AA-Briefcase and AutomationBench-AA for work tasks, FrontierCode 2.0 and Terminal-Bench 4.0 for coding, and Terminal-Bench-Science and CritPt v2 for science.
The benchmark maker says 60% of v5 uses private test sets or held-out answers, intended to reduce the chance that models have memorized test material. Many evaluations involve longer agent tasks, such as working in a terminal, producing reports or fixing code. Those design choices make the index relevant to multi-step work, but they also mean its ranking depends on the chosen tests, model settings and scoring rules.
The timing is notable: the source says Claude Opus 5.5 led the newly rebuilt leaderboard on October 28, before Carbon moved into first place the following day. The available information describes the new result but does not establish what caused the ranking to change. The index should be read as a dated snapshot rather than a permanent ordering.
Release Timing and Benchmark Limits
Google has not announced a public release date for Carbon. The reported rollout plan is limited access for cyber defenders first, then trusted partners, followed by developers, businesses and consumers; the timing and eligibility for each stage remain unspecified.
The source reports introductory API pricing of $2 per million input tokens and $10 per million output tokens, rising later to $4 and $20, respectively. It does not identify when those rates begin or change, or provide enough detail here to establish how the quoted per-task benchmark cost translates to a customer’s workload. Actual costs can depend on task length, settings and usage.
The full component results and methodology matter when interpreting a one-point lead. The source describes several areas where Carbon loses, but does not provide complete results for every test in the material supplied. The ranking also cannot establish which model will perform best on an individual organization’s data or applications.
Google’s Staged Access Rollout
The next reported step is access for a subset of cyber defenders, followed by trusted partners. Google’s stated plan then moves to developers, businesses and consumers, but no dates were provided. Until access widens, independent customers cannot assess Carbon directly against their own tasks or verify whether the benchmark’s relative strengths carry over to their work.
As availability expands, relevant points to watch include the final API pricing and release schedule, independent evaluations beyond the nine index tests, and results on practical workloads such as coding, document handling and business processes. For now, the confirmed development is a new benchmark leader—not a public product launch, nor evidence that Carbon is best for every use.
Key Questions
How far ahead is Gemini 4 Carbon?
Carbon scores 59 points on the Artificial Analysis Intelligence Index v5, one point ahead of Claude Opus 5.5 at 58 and two ahead of GPT-6 Astra at 57.
Is Gemini 4 Carbon available to the public?
No. The source says Google is initially giving access to a subset of cyber defenders, with broader groups planned afterward. No public release date is stated.
Does the top ranking mean Carbon is best at every task?
No. The index combines nine tests, and Carbon leads four. The reported results put other models ahead on office-style knowledge work, coding, physics research, terminal tasks and document reasoning.
Why does Carbon score well on the omniscience test?
Artificial Analysis reports that the test penalizes wrong answers. Carbon has lower reported accuracy than some rivals on the test, but an 11% hallucination rate, compared with about 50% for Claude Opus 5.5 and GPT-6 Astra. That lower rate of incorrect answers can improve its score under the test’s rules.
What will Carbon’s API cost?
The source lists introductory rates of $2 per million input tokens and $10 per million output tokens, later rising to $4 and $20. It does not provide dates for those pricing stages.
Source: ThorstenMeyerAI.com
