Google's Gemini 4 Carbon, which Google announced on October 28 as its strongest model, ranks first on the Artificial Analysis Intelligence Index v5 with a score of 59. The developers and businesses who buy models cannot use it. Google released it only to "a subset of cyber defenders" through its Fairwind Program, and promised developers, businesses and consumers access "soon", without a date.
It is the second time in a month. Gemini 4 Argon, announced on September 30, went to Fairwind defenders first and reached developers 13 days later, on October 13. As of November 1, Carbon has been gated for four days, by a SaasBuzz count of Google's own dates, and the Gemini API's model list and changelog do not list it. For a company whose investors just wiped about $500 billion off its value over AI spending, the model it ranks first is, for now, one that no customer can buy off the shelf.
Two launches, two gates
Google's posts set out the same sequence twice. The September 30 Argon post said the model was "rolling out to a set of trusted cyber defenders through our Fairwind Program" and that Google would release to developers, enterprises and consumers "starting with paid API customers and Google AI Ultra subscribers". On October 13, a second post said Argon was "now globally available today for developers and enterprises", and the Gemini API changelog lists Argon as generally available that day. Fairwind partners kept something extra: access to Argon "without cyber guardrails".
Model | Defenders | Everyone | Days gated |
|---|---|---|---|
Gemini 4 Argon | Sep. 30 | Oct. 13 | 13 |
Gemini 4 Carbon | Oct. 28 | Not yet | 4 so far |
Source: SaasBuzz analysis of Google's blog posts and the Gemini API changelog, as of November 1, 2026.
Carbon's post gives the reason in one passage: Google is "actively engaged in the U.S. government's voluntary process for pre-release model testing", and is "continuing to gather feedback and iterate on guardrails before making Carbon available to developers, businesses, and consumers via our product surfaces as soon as possible." The Fairwind Program page says Google prioritizes "governments and partners most critical to societal resilience", such as national cyber authorities and critical infrastructure operators, and tells applicants to contact their Google Cloud account team.
Sundar Pichai, chief executive of Google and Alphabet, framed the gate as a deliberate repeat, writing on X that "As we did with Argon, we're going to roll it out gradually + responsibly".
Post on X by @sundarpichai
I did say we'd be releasing our models periodically...introducing Gemini 4 Carbon, our next frontier model. Following Argon's release just two weeks ago, Carbon shows the pace at which we’re pushing the frontier, it ranks #1 on the recently updated v5.0 of @ArtificialAnlys Intelligence Index at 59. It also leads across key cybersecurity evaluations scoring 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation, and 71.0% on CWE-bench for patching. We’re seeing significant improvements in long-horizon tasks even compared to Argon, and we’ve strengthened frontier safeguards. As we did with Argon, we’re going to roll it out gradually + responsibly. Some trusted cyber defenders will get access starting this week. We'll be expanding it to more partners through our Fairwind Program, ahead of wider availability soon, so hang tight!
What the top spot is worth
The lead is narrow. On the Artificial Analysis Intelligence Index v5, Carbon scores 59, Anthropic's Claude Opus 5.5 58 and OpenAI's GPT-6 Astra 57. Unrounded, from the values the page publishes, Carbon's lead over Opus 5.5 is 1.02 points. Artificial Analysis also puts Carbon 9 points above Argon, the Gemini model developers can actually call.
Model | Index v5 | Cost per task |
|---|---|---|
Gemini 4 Carbon | 59 | $7.76 |
Claude Opus 5.5 | 58 | $10.04 |
GPT-6 Astra | 57 | $6.25 |
Gemini 4 Argon | About 50 | $7.68 |
Source: Artificial Analysis Intelligence Index v5 and its October 28 results thread; cost to run the index's tasks at each model's standard price, best-scoring setting. Argon's score is Carbon's less the 9 points Artificial Analysis reported.
Divided out, Carbon costs about 13.1 cents per index point per task, against 17.2 cents for Opus 5.5 and 11.0 cents for GPT-6 Astra, by a SaasBuzz calculation from those figures. So the model that tops the index is cheaper to run than its nearest rival and dearer than the third.
Google's own page settles the price that buyers will eventually pay. Carbon "will launch at the same introductory price as Argon at $2 per million input tokens and $10 per million output tokens", with cached input 95% off, and "After the introductory period expires, the price of $4 per 1M input tokens and $20 per 1M output tokens will apply." The introductory rate is half the standard one. Google has not said when Carbon's introductory period ends; for Argon, the footnote sets it at November 30, 2026.
What the industry is saying
Google's own people have stuck to one line: defenders first, everyone else next. Josh Woodward, whose X profile lists him as a Google vice president for Google Labs, the Gemini app and AI Studio, put the sequence plainly on launch day.
Post on X by @joshwoodward
2 weeks ago: Gemini 4 Argon Today: Gemini 4 Carbon The frontier is fast, and the internal vibes on this model are good. Rolling out to cyber defenders first, then more widely available as the next step!
The independent rankers confirmed the score and the gate. Artificial Analysis, which runs the index Google cites, wrote that Carbon "is being rolled out to selected users and is not yet publicly available", as Argon was. Its thread also carried caveats Google's post did not: Carbon is "not leading in all these areas", ranks behind Claude Opus 5.5 on its coding agent index, and its hallucination rate on AA-Omniscience rose to 11% from Argon's 4%, though that is still far below the 50% it measured for Opus 5.5 and GPT-6 Astra.
Post on X by @ArtificialAnlys
Google’s Gemini 4 Carbon is the new #1 on the Artificial Analysis Intelligence Index v5, and second only to Claude Opus 5.5 on the Artificial Analysis Coding Agent Index Gemini 4 Carbon scores 59 on Intelligence Index v5, driven by leading performance in complex reasoning, knowledge, and scientific tasks. It also has large gains in coding and knowledge work, particularly when compared to other Gemini models, but it is not leading in all these areas. As with launch of Gemini 4 Argon, Gemini 4 Carbon is being rolled out to selected users and is not yet publicly available. Google is offering a 50% launch discount, with no confirmed end date. All our charts and figures use standard pricing for both Gemini 4 Carbon and Argon. Key results for Gemini 4 Carbon with high reasoning: ➤ Gemini 4 Carbon leads the Intelligence Index v5, with the #1 score in 4 of the 9 Index evaluations: HLE-Diamond, Terminal-Bench-Science, AA-Omniscience, AutomationBench-AA. Carbon makes substantial gains elsewhere but is not leading in coding and knowledge work ➤ Behind only Claude Opus 5.5 on coding agents: With Antigravity CLI, Gemini 4 Carbon (high) scores 59 only 3 points behind Claude Code with Claude Opus 5.5 (max, 62) on the Artificial Analysis Coding Agent Index. Carbon ranks second on both Terminal-Bench 4.0 (64%) and FrontierCode v1.1 (51%), behind Claude Opus 5.5 (max) at 56%. Google has historically underperformed in this area and so this improvement stands out ➤ Improved agentic knowledge work, but not yet leading: On AA-Briefcase, Gemini 4 Carbon (high) scores 1609 Elo, a 126-point improvement over Gemini 4 Argon (high, 1483), but still behind models including Grok 4.7 (xhigh, 1638), Muse Wonder (max, 1659) and Kimi K3.1 Pro (1678). The largest gain comes from Presentation, up 293 Elo to 1598, alongside a 120-point improvement in Analytical Quality to 1671 ➤ Lower Cost per Task than Claude Opus 5.5 for comparable overall intelligence: Gemini 4 Carbon costs $7.76 per Intelligence Index task, ~23% less than Claude Opus 5.5 (max, $10.04) but ~1.2x GPT-6 Astra (max, $6.25). It uses 123k output tokens per task, 29% fewer than Claude Opus 5.5 (max, 173k). At the same $4/$20 standard pricing, its Cost per Task is in line with Gemini 4 Argon ($7.68), while scoring 9 points higher on the Intelligence Index ➤ Leading AA-Omniscience v1.1 score, with low hallucination: Gemini 4 Carbon scores 54 on our recently updated AA-Omniscience v1.1, 6 points ahead of Claude Opus 5.5 (max, 48) and 8 points ahead of Gemini 4 Argon (46). Accuracy (% correct) rises 10 points over Argon to 58%, while its hallucination rate rises from 4% to 11%. This is still the second lowest of any model, behind only Argon, compared with 50% for both Claude Opus 5.5 (max) and GPT-6 Astra (max) Key model details: ➤ Context Window: 1M tokens ➤ Multimodality: Text, image, video, and speech input, with text output ➤ Pricing: $4/$20 per 1M input/output tokens, the same as Gemini 4 Argon’s standard pricing, with a 50% discount. Cached input tokens receive a 95% discount ($0.20 per 1M) Congratulations to the @GoogleDeepMind team!
Arena, which ranks models by the votes of its users, placed Carbon first in its text ranking and second, behind Claude Opus 5.5, in its agent ranking. Its rankings do not give anyone a way to buy it.
Post on X by @arena
Exciting news: Gemini 4 Carbon (High) from @GoogleDeepMind just made a huge debut on Arena, landing #1 in Text Arena, #2 in Agent Arena, and #4 in Code Arena: WebDev. Back at the top, Gemini 4 Carbon delivers frontier intelligence and reshapes the Pareto frontier across three arenas. In Agent Arena, Gemini 4 Carbon ranks #2 with a +16.2% net improvement across +21.9K real-world agentic sessions from our global community, surpassing top models GPT-6 Astra and Fable-5.1. At $1.80 median cost per task, Gemini 4 Carbon (High) costs 72% less than the #1 ranked Claude Opus 5.5 (Max). Its standouts by key signals: - #1 Steerability (+18.6%): can the model course-correct when you push back? - #2 Praise vs Complaint (+37.9%): implicit sentiment in users' reactions - #3 Confirmed Success (+18.7%): explicit “yes, that worked” from users In Code Arena: WebDev, Carbon scored 1768 pts landing #4: - Just 2 pts below #3 Claude Sonnet 5.5 (xHigh), at 1770 pts - 89 pts above #11 Gemini 4 Argon (High), at 1679 pts By category, it ranks #3 in Gaming and Simulations, and #5 in Brand & Marketing. In Text Arena, Gemini 4 Carbon (High) is the new #1 with 1528 pts, now ahead of Gemini 4 Argon (High) at #2 with 1520 pts. Congrats to the @GoogleDeepMind team on this huge milestone!
Google's own launch posts, from Pichai's down, said who could use Carbon in the announcement itself, and none gave a date.
Is this the industry's pattern, or Google's?
The narrative that frontier models now arrive gated holds, with differences in how much is held back and for how long. Every leading lab now holds its most cyber-capable access behind vetting. OpenAI, like Google, released its September flagship to vetted organizations first; Anthropic released Opus 5.5 to everyone and restricted its cyber use.
Anthropic released Claude Opus 5.5 on September 22 on "all platforms, including Amazon Web Services, Google Cloud, and Microsoft Azure", at $4 and $20 per million tokens. It gated the use instead: because Opus 5.5 is comparable to its Claude Mythos 5.1 in cybersecurity, Anthropic said "most cybersecurity tasks will be re-routed to Opus 4.8" for general users, and said it would "soon" expand its Cyber Verification Program to Opus 5.5 for verified defenders; the program "will include three tiers for increasingly permissive trusted access", including access to its Claude Mythos models. OpenAI released GPT-6 Astra on September 3 "initially for those in its Trusted Access Program", The Register reported, with paid ChatGPT subscribers to follow within days; GPT-6 Astra is now listed in OpenAI's API, and OpenAI runs Daybreak Blue, its vetted program for cybersecurity work.
The strongest counterargument is that Google's gate is short and the difference is cosmetic. Argon's lasted 13 days, and Anthropic's model also re-routes most cybersecurity tasks to an older model for unverified users. That answer works for a buyer who can wait two weeks. It does not work for one who has to plan a quarter: Google gave Argon's general release a sequence, not a date, and it has done the same for Carbon. Two weeks is a pattern of one.
What it means for developers and buyers
Procurement cannot start from the leaderboard. The model ranked first on Artificial Analysis is not on Google's public API, and Fairwind access runs through a Google Cloud account team and a vetting process aimed at governments and critical infrastructure.
Argon is what Google sells today, at the same introductory $2 and $10 per million tokens, with that rate scheduled to end on November 30. Teams pricing a Gemini build should model the $4 and $20 standard rate, which Carbon will also carry.
The one-point lead is within the range where workload matters more than rank. Artificial Analysis's own figures put Opus 5.5 ahead on coding agents and GPT-6 Astra cheaper per task, and both can be bought now.
Security teams that qualify have a reason to apply: Google says Carbon set new highs on SEC-bench Pro, at 88.2%, and ExploitGym, at 54.8%, on its own testing.
What would change the picture is a date. If Carbon reaches the Gemini API on roughly Argon's timetable, around November 10, defenders-first is a two-week head start for security teams. If it does not, the model Google ranks as its best stays one that most of its customers cannot buy, and buyers will keep choosing between Argon and models from Anthropic and OpenAI that are on sale today.
Method
Dates come from Google's blog posts of September 30, October 13 and October 28 and the Gemini API changelog, checked on November 1, 2026. Days gated run from the defenders-only announcement to general availability, or to November 1 for Carbon. Index scores and costs per task are Artificial Analysis's, from its index page and its October 28 thread; the cents-per-point figure divides each model's cost per task by its unrounded score. The cybersecurity benchmark results are Google's own and have not been independently reproduced; the coding-agent and Arena rankings are Artificial Analysis's and Arena's. The calculation is in the package's calc folder.
