🔍 Read the full analysis: The Ranking Gives Gemini 4 Carbon A One-Point AI Lead. What Comes Next? 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’s Intelligence Index v5, one point ahead of Claude Opus 5.5 and two ahead of GPT-6 Astra. It leads on four of nine tests, but trails competitors on several other tasks and is not yet publicly available; Google has not given a rollout date.
Google’s Gemini 4 Carbon has taken the top position 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 is a narrow lead on a specific set of nine tests, not a general measure of which AI model is best, and Carbon is not yet available to the public.
Artificial Analysis launched version 5 of its index on October 27, and the source report says Carbon moved into first place the following day. In the reported high-reasoning configuration, Carbon scored 59; Claude Opus 5.5 at maximum reasoning scored 58, GPT-6 Astra at maximum reasoning scored 57, GPT-6.1 Sol scored 55, and Google’s previous Gemini 4 Argon scored 50.
Carbon placed first on four of the nine tests: HLE-Diamond, Terminal-Bench-Science, AA-Omniscience and AutomationBench-AA. Its reported results include 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. It also scored 78% on AutomationBench-AA, fractionally ahead of Argon, according to the report.
It placed behind other models on the remaining five tests. Claude Sonnet 5.5 and Claude Opus 5.5 led office-style knowledge work; Claude Opus 5.5 led the real-world coding test; GPT-6 Astra led the physics research test; Kimi K3.1 Pro led terminal tasks; and Meta’s Muse Wonder Full led document reasoning. The source also reports a cost of $7.76 per index task at standard prices. Its comparisons put that below Claude Opus 5.5 at $10.04, but above GPT-6 Astra at $6.24 and GPT-6.1 Sol at $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 One-Point Lead Has Limits
The ranking gives companies a useful comparison, but the narrow margin and uneven test results argue against choosing a model by its overall place alone. Carbon’s 59-to-58 lead does not mean it will outperform Claude on office work or coding, or GPT-6 Astra on physics research. The relevant result depends on the work a company needs done and the model settings used.
Cost also depends on workload. The reported $7.76 per index task is a benchmark-specific estimate, not a universal price for every user or task. The source lists introductory API pricing of $2 per million input tokens and $10 per million output tokens, later rising to $4 and $20 respectively. Buyers will need to test their own tasks and account for both performance and spending before switching.
For individual users, there is no immediate change because Carbon has not opened to the public. If and when access expands, the report expects its advantages to show most on long, difficult tasks, rather than routine chat. That is a forecast in the source, not a demonstrated outcome for general users.
Products Worth Considering
DJI Osmo Pocket 3 Creator Combo Vlogging Camera with Mic 2, 1 CMOS, 4K/120fps, 3-Axis Stabilization, 128GB Card
Bose QuietComfort Wireless Bluetooth Headphones with Active Noise Cancellation, Mic, USB‑C Charging, Deep Bass, 24‑Hour Battery – White Smok
Amazon Kindle Paperwhite (12th Generation, 2024), 16 GB, 7-inch Display, Jade
As an affiliate, we earn on qualifying purchases.
What Version Five Measures
Artificial Analysis describes its Intelligence Index as a combined score across nine tests, grouped into four equally weighted areas: knowledge work, coding, science and general capability. The source says the index launched October 27 and that 60% of its tests use private test sets or held-out answers, intended to reduce the chance that models have memorized the questions.
Many tasks involve longer agent-style work, such as operating in a terminal, producing reports or modifying code. That is different from asking a model a short question. The result therefore reflects performance on this particular collection of tasks, at the settings tested. It does not establish a universal ranking across every use case.
The report’s AA-Omniscience result illustrates how test design affects scores. Carbon earned 54 points on a measure that penalizes wrong answers, six ahead of Claude Opus 5.5. Yet Carbon’s plain accuracy was reported at 58%, below Claude Fable 5.1 at 68%, Claude Opus 5.5 at 66%, and GPT-6.1 Sol and GPT-6 Astra at 62%. Artificial Analysis’s reported hallucination rate—wrong answers as a share of questions the model did not answer correctly—was 11% for Carbon, compared with about 50% for Claude Opus 5.5 and GPT-6 Astra. The higher score reflects the penalty for incorrect answers, not the most correct answers overall.
Access and Real-World Results
Google has not announced a public release date for Gemini 4 Carbon. The source says Google plans to give access first to a subset of cyber defenders, then trusted partners, followed by developers, businesses and consumers; the timing of each stage is unspecified.
The benchmark report does not establish how Carbon will perform across every company’s workflows, or whether the listed index-task costs will match costs for production use. The supplied source material also cuts off during its discussion of the coding comparison, so it does not provide the full accompanying explanation. Readers should treat the reported score and pricing as specific to the stated benchmark and source, rather than as independently verified results for all settings.
Google’s Staged Access Plan
The next stated milestone is limited access for a subset of cyber defenders, followed by trusted partners. Google’s reported plan then extends to developers, businesses and consumers, but the company has not provided dates for those stages in the source material.
For organizations considering Carbon, the practical next step is to compare it with alternatives on representative tasks, including accuracy, refusal behavior, completion time and cost per finished task. Until access widens and those evaluations are possible, the leaderboard indicates where Carbon performed well in Artificial Analysis’s tests—not whether it is the right choice for a particular deployment.
Key Questions
How far ahead is Gemini 4 Carbon?
It 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.
Did Carbon win every benchmark in the index?
No. The source report says Carbon placed first on four of nine tests and behind other models on the other five, including office-style work and real-world coding.
Can the public use Gemini 4 Carbon now?
No public release is reported. Google is said to be starting with a subset of cyber defenders, but no rollout dates are provided.
Does the top index score mean Carbon knows more?
Not on every measure. The report says several Claude and GPT models had higher plain accuracy on AA-Omniscience; Carbon scored well there because the test penalizes wrong answers and its reported hallucination rate was lower.
What does the reported task cost represent?
The source lists $7.76 per index task at standard prices. It is a benchmark-specific estimate and should not be read as a fixed cost for every real-world task.
Source: ThorstenMeyerAI.com
