🔍 Read the full analysis: One Point Separates Gemini 4 Carbon From The Rest Of The AI Ranking on ThorstenMeyerAI.com
Get the latest gadgets 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. The ranking covers nine tests, and Carbon leads four while trailing on five; Google has not made the model publicly available and has not announced a rollout date.
Gemini 4 Carbon ranks first on Artificial Analysis’ Intelligence Index v5 with 59 points, one point ahead of Claude Opus 5.5 and two ahead of GPT-6 Astra. The result puts Google’s model at the top of a nine-test benchmark, but Google is initially limiting access to a subset of cyber defenders, and the score does not mean Carbon leads every task or is the best fit for every user.
The index, rebuilt and launched by independent benchmarking company Artificial Analysis on October 27, combines results from nine tests into a single score. The supplied ranking lists Claude Opus 5.5 at 58, GPT-6 Astra at 57, GPT-6.1 Sol at 55 and Google’s earlier Gemini 4 Argon at 50. Carbon’s score is for its high-reasoning setting; the listed competitor scores use the settings named in the ranking.
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, which cover office-style work, coding, physics research, terminal tasks and document reasoning. For example, Claude Opus 5.5 scores 46% on FrontierCode 2.0 to Carbon’s 41%, while Claude Sonnet 5.5 and Opus 5.5 lead Carbon on the AA-Briefcase office-work test.
The benchmark source puts Carbon’s cost at $7.76 per index task at standard prices. That is below the listed $10.04 for Claude Opus 5.5, but above GPT-6 Astra’s $6.24 and GPT-6.1 Sol’s $1.27. Google’s stated introductory API price is $2 per million input tokens and $10 per million output tokens, later rising to $4 and $20 respectively, according to the source material.
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 Practical Trade-Offs
The one-point margin matters because a headline ranking can look more decisive than the underlying results. Carbon’s 59-to-58 advantage reflects performance across this particular mix of tests and weightings. It is not evidence that Carbon outperforms Claude or GPT models on every job. Companies comparing systems will need to weigh their own tasks, response quality, reliability and cost per completed task, rather than choosing solely by overall rank.
The results also illustrate how a benchmark’s scoring rules shape what a model’s lead means. Artificial Analysis’ AA-Omniscience test penalizes incorrect answers. Carbon scores 54 on that measure, six points ahead of Claude Opus 5.5, even though several models score higher on plain answer accuracy. The source reports Carbon at 58% accuracy and an 11% hallucination rate, compared with about 50% for Claude Opus 5.5 and GPT-6 Astra on the test’s hallucination measure. That suggests Carbon’s advantage there is associated with fewer wrong answers, not the highest number of correct answers.
For ordinary users, the immediate effect is limited: Carbon is not publicly available. For developers and businesses, availability and price may matter as much as the leaderboard result. The reported API prices are not the same as the index’s per-task cost, and actual spending will depend on how a system is used.
Products Worth Considering
LG UltraGear 32GX850A-B 32-Inch 4K UHD Glossy OLED Gaming Monitor
Apple iPad Air (2025) 13‑inch Wi‑Fi 256 GB – Space Gray (M3 chip)
Apple iPad Air (4th Generation, 10.9-inch, Wi-Fi, 256GB) – Space Gray
As an affiliate, we earn on qualifying purchases.
How Artificial Analysis Scores Models
Artificial Analysis describes its Intelligence Index as a combined score from nine tests arranged in four equally weighted blocks: knowledge work, coding, science and general capability. Within those blocks, tests have different individual weights. For example, AA-Briefcase and FrontierCode 2.0 each account for 15% of the overall index, while several other tests account for 5% or 10%.
Version 5 launched on October 27, 2026. According to the source material, 60% of the index now uses private test sets or held-out answers. Many tasks require models to work through longer processes, such as using terminal tools, producing reports or fixing code. That approach is intended to test more than short question-and-answer performance, but it also means the final score remains tied to the benchmark’s selected tasks and scoring design.
The reported ranking compares particular model settings, not every configuration or product experience available from each company. Separate tests can produce different leaders: the source says Claude Opus 5.5 leads the office-work and real-world coding comparisons, while other models top physics research, terminal tasks and document reasoning. The ranking is best read as a snapshot across specified evaluations, not a universal judgment about AI capability.
Access, Rollout and Benchmark Limits
Google has not provided a public release date in the supplied material. The reported rollout plan starts with a subset of cyber defenders, followed by trusted partners and then developers, businesses and consumers, but the timing and eligibility for each stage remain unspecified. It is also unclear when users outside the initial group will be able to try Carbon.
The source does not provide full results for every one of the nine tests or detailed information about the costs behind every model’s runs. Its figures are a reported snapshot, and the ranking may change as evaluations or model versions change. The score also cannot establish how Carbon performs on a particular company’s data or workflow; those results would require testing in that setting.
Price comparisons need care: the source gives both a per-index-task estimate and an API token price, which measure different things. It does not provide enough detail here to calculate a typical customer’s total bill. Nor does the benchmark alone establish whether the reported differences will be noticeable in everyday use.
Google’s Staged Release Ahead
The next development to watch is Google’s move from the initial cyber-defender group to trusted partners and broader developer access. The company has not supplied dates for those stages in the source material, so a consumer launch cannot be forecast from the ranking report alone.
As access expands, independent evaluations and customer trials could clarify how Carbon performs outside the nine index tests, what its real operating costs look like and whether its strengths fit specific workloads. Until then, the confirmed news is a narrow lead on one benchmark, alongside limited availability and mixed results across individual tasks.
Key Questions
What is Gemini 4 Carbon’s score?
It scored 59 points on Artificial Analysis’ Intelligence Index v5, according to the supplied October 29 report.
How far ahead is Carbon?
Carbon is one point ahead of Claude Opus 5.5, listed at 58, and two points ahead of GPT-6 Astra, listed at 57.
Does Carbon lead every test in the index?
No. The report says Carbon leads four of nine tests and trails on the other five, including office-style knowledge work and real-world coding.
Can the public use Gemini 4 Carbon now?
Not according to the supplied report. Google is starting with a subset of cyber defenders; no public release date is given.
Does the index prove Carbon is the best AI model?
No. It ranks model performance on a specific set of nine tests with defined weights. The result does not establish that Carbon is best for every task, user or business workflow.
Source: ThorstenMeyerAI.com
