🔍 Read the full analysis: What A One-Point Margin Means For Gemini 4 Carbon’s AI Model Ranking on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Gemini 4 Carbon ranks first on Artificial Analysis’s Intelligence Index v5 with 59 points, one ahead of Claude Opus 5.5 and two ahead of GPT-6 Astra. It leads on four of nine tests, but the score is not a broad verdict on model quality, and Carbon is not yet publicly available.
Google’s Gemini 4 Carbon ranks first on Artificial Analysis’s newly rebuilt Intelligence Index v5, scoring 59 points—one ahead of Claude Opus 5.5 and two ahead of GPT-6 Astra. The result gives Google the top position on this particular nine-test benchmark, but Carbon is not yet publicly available and does not lead every test in the index.
Artificial Analysis launched version 5 of its index on October 27, then ranked Carbon first the following day, according to the October 29 source report. The listed 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, scores 50.
Carbon places first on four of the nine tests: HLE-Diamond, Terminal-Bench-Science, AA-Omniscience and AutomationBench-AA. It does not lead the other five. Claude models perform better on office-style knowledge work, Claude Opus 5.5 leads the cited real-world coding test, GPT-6 Astra leads the physics research test, Kimi K3.1 Pro leads a terminal-task test, and Meta’s Muse Wonder Full leads document reasoning.
The source reports an estimated $7.76 per index task at standard prices. That is below the reported $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 introductory API price is listed as $2 per million input tokens and $10 per million output tokens, later rising to $4 and $20 respectively. These are reported prices; actual costs depend on usage and task requirements.
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
What a One-Point Lead Can Tell Buyers
The score makes Carbon a leading option to evaluate, not an automatic choice for every workload. A one-point margin is small relative to the index’s overall scale, and the models trade places across task categories. A company doing business-process automation or science workflows may value Carbon’s reported strengths differently from one whose main need is coding or document analysis.
Cost also changes the comparison. The reported index-task estimate puts Carbon below Claude Opus 5.5 but above GPT-6 Astra and far above GPT-6.1 Sol. That does not establish which model is cheapest for a particular company: billing can vary with input and output volume, retries, tools and how often a model completes a task successfully. Buyers need to compare cost per completed task, not just a benchmark rank or token price.
For individual users, there is no immediate product change because Carbon has not been released publicly. If and when it becomes generally available, the source expects its strongest differences to show up on longer, demanding tasks rather than routine chat. That is an expectation, not a demonstrated outcome for everyday users.
Popular right now

De'Longhi Magnifica Start Automatic Espresso Machine with Milk Frother, Silver

Breville Barista Express Manual Espresso Machine with Built-In Grinder, Brushed Stainless Steel

Samsung 55-Inch QN70H Neo QLED 4K Smart TV (2026)
As an affiliate, we earn on qualifying purchases.
How Artificial Analysis Built Version Five
Artificial Analysis describes its index as a combined score from nine tests, arranged into four equally weighted blocks: knowledge work, coding, science and general capability. The reported weights include 15% each for AA-Briefcase, FrontierCode 2.0, Terminal-Bench-Science and AA-Omniscience, with the remaining tests making up the rest of each block. The index therefore reflects both test selection and weighting, rather than a single measure of general intelligence.
The source says 60% of version 5 uses private test sets or held-out answers. Many tasks also require models to act over longer workflows, such as using a terminal, producing reports or fixing code. Those choices are intended to test practical work and reduce the advantage of memorized public answers. They also make the tests more involved and costly to run.
One result illustrates how a test’s scoring rules affect rankings. Carbon scores 58% on plain accuracy in AA-Omniscience, below several Claude and GPT models, but leads the test’s overall score because it is less likely to give a wrong answer when uncertain. The source reports an 11% hallucination rate for Carbon, compared with about 50% for Claude Opus 5.5 and GPT-6 Astra under that test’s definition. Its result is therefore a win on a benchmark that penalizes wrong answers, not evidence that Carbon knows more than those models.
What the Ranking Leaves Unsettled
The available source does not provide the full Artificial Analysis methodology, underlying results for every test, or enough detail to independently assess how the overall score responds to small changes in weighting. The reported one-point gap should be read as the difference in this index’s scores, not as proof of a decisive or statistically meaningful advantage across all uses.
It is also unclear when Carbon will reach developers, businesses and consumers. Google’s stated sequence is a limited release to cyber defenders, then trusted partners, followed by broader groups, but the source gives no rollout dates. The source text also ends partway through a sentence about the coding comparison, leaving the explanation incomplete. The ranking alone does not establish real-world performance, reliability, privacy terms or total cost for any specific buyer.
Watch for Access and Independent Testing
The next visible milestone is access for the initial group of cyber defenders, followed by trusted partners, according to the source. Google has not supplied a date for those steps or for wider release. Further reporting should check whether the rollout follows that sequence, what product and API limits apply, and whether the announced introductory prices change as described.
For businesses considering a switch, the practical next step is to test Carbon on representative work alongside alternatives, using the same task criteria and measuring completion quality, error rates, latency and total cost. Future benchmark results may also change the leaderboard as more models and tests are added. Until release details and task-specific results are available, the confirmed takeaway is limited: Carbon leads this version of this index by one point, while the best model for a given user or company remains dependent on the work being done.
Key Questions
What does Gemini 4 Carbon’s one-point lead mean?
It means Carbon scored 59 on Artificial Analysis’s Intelligence Index v5, one point above Claude Opus 5.5. It does not mean Carbon is best on every task or that it will outperform competitors in every real-world use.
Which tests does Carbon lead?
The source report says Carbon ranks first on HLE-Diamond, Terminal-Bench-Science, AA-Omniscience and AutomationBench-AA. It loses the other five tests to different models.
Is Gemini 4 Carbon available to the public?
No public release is reported. Google is starting with a subset of cyber defenders, then plans to expand to trusted partners and later to developers, businesses and consumers. No dates are provided.
Is Carbon the cheapest leading model?
Not according to the source’s index-task estimates. Carbon is reported at $7.76 per task, less than Claude Opus 5.5’s $10.04 but more than GPT-6 Astra’s $6.24 and GPT-6.1 Sol’s $1.27. Costs for a particular workload can differ.
Source: ThorstenMeyerAI.com
