Verified · Nov 11, 2026
Independently verifiedGemini 4 Carbon is now available — the model Google calls its strongest lands on paid API tiers and AI Ultra, at a $2/$10 intro price that steps up after November 30
2 sourcesGoogle's official post (Tuesday, November 10; opened and read in full this run) announces: "Today, we’re making Gemini 4 Carbon available to developers, enterprises, scientists, and consumers." and describes Carbon as "our strongest model with frontier performance in cybersecurity capabilities and long-horizon agentic workflows across complex knowledge work, software engineering, and scientific reasoning". The sequencing per the post: "Two weeks ago, we released Carbon to a set of cybersecurity defenders via our Fairwind Program."; "Following our recent launch of Gemini 4 Argon, Carbon pushes the frontier even further."; and the division of labor — "Argon remains the ideal choice for the majority of software engineering and knowledge work, while Carbon is built for tasks that test the limits of a model’s intelligence like complex, multi-step coding and debugging and open-ended scientific research and discovery", "using more tokens than Argon to explore thoroughly, reason deeply, and run extended verification loops". Benchmarks per the post: "It ranks #1 on CursorBench 4.0, Perplexity’s WANDR, and Zapier’s AutomationBench at more than 2X lower cost per task compared to the next best external model." and "Carbon ranks #2 while being more than 6X lower in average cost per task compared to the top performing model." on the new Multimodal Knowledge Agent benchmark Google says it "partnered with Artificial Analysis to launch". The agent feature is a preview: Google is "previewing /swarm in Google Antigravity, turning a single prompt into a coordinated team of agents you can track in real time". Availability per the post: "Gemini 4 Carbon is available via the API in Google AI Studio and Gemini for business for all paying developers and enterprises" at the same introductory price as Gemini 4 Argon ($2 per million input tokens and $10 per million output tokens), and "It is also rolling out to all Google AI Ultra subscribers in Google Antigravity, the Gemini app, and AI Mode in Search." The post's footnote sets the clock: "After the introductory period expires on November 30, 2026, the price of $4 per 1M input tokens and $20 per 1M output tokens will apply." — applying to "the Carbon model intended for public access only". The Verge's report (same day) relays the rollout — "Google is rolling out its latest AI model, Gemini 4 Carbon, to the general public after releasing it to a select group of “cybersecurity defenders” two weeks ago." — and adds its own context: the intro price is "around a fifth the cost of OpenAI’s GPT-6 Astra and a quarter the cost of Anthropic’s Fable 5.2 model, which launched yesterday", and "It is not clear when — or if — Carbon will be available to lower tier subscribers or free users."
Why now
The post landed Tuesday, November 10, with a same-day relay wave (The Verge, Engadget, 9to5Google, VentureBeat, and more per the Google News RSS relay set saved this run). For creators, this is a checkable-numbers model release with a pricing clock built in: Google's own footnote puts the intro price's expiry on the page (November 30, 2026), so a pricing explainer published this week can read the footnote aloud before the step-up date. The consumer angle is gated and the gates are the story: paying developers and enterprises via the API, Google AI Ultra subscribers in the Gemini app, AI Mode in Search, and Google Antigravity per the post, /swarm a preview — and The Verge's gap line about lower tiers and free users is the open question creators can legitimately raise on screen.
Why it is worth publishing
A same-day official-plus-relay model card where every number carries its attribution and its bound ('more than 2X', 'more than 6X'), the availability gates stay visible (paying developers and enterprises, AI Ultra subscribers, /swarm in preview), and the pricing timeline is checkable against the post's own footnote rather than any outlet's arithmetic.
Evidence basis
Two opened sources: Google's official post (November 10; raw HTML 407,351 bytes; JSON-LD datePublished 2026-11-10T16:00:00+00:00; November 10 = Tuesday, calendar-verified once this run) carries every announcement fact, every benchmark and safety number, and the pricing footnote; The Verge (November 10; raw HTML 390,947 bytes; byline Robert Hart) is the opened media relay adding the rival-cost comparison, the cherry-picking caution, and the lower-tier availability gap. Calendar computations for this run, computed once with the system calendar: 2026-11-10 = Tuesday; 2026-11-11 = Wednesday.
“The model Google calls its strongest is now open to paying developers — at a $2 and $10 intro price that Google's own footnote replaces with $4 and $20 after November 30.”
Angle
Cover it as a checkable-numbers release: Google's own sentences on screen with every benchmark and safety number carrying its 'per Google' attribution, the pricing timeline ($2/$10 intro, $4/$20 after November 30 per the post's footnote) as the practical news, and Argon-versus-Carbon positioning explained in the post's own terms.
Format
Long-form explainer
Demo idea
A pricing-timeline card: the post's intro price ($2 per million input tokens and $10 per million output tokens) on one row and the footnote's post-November-30 price ($4 per 1M input tokens and $20 per 1M output tokens) on the next, with the footnote sentence itself on screen — the one pricing claim on this card that carries its own expiry date in Google's own text.
Platform notes
The trust moat is vendor-claims discipline: every benchmark and safety number is Google's own claim about its own model, no opened source independently benchmarked any of it, and the 'more than' bounds never drop ('more than 2X lower cost per task' is not '2X lower cost'). The rival-cost comparison (GPT-6 Astra, Fable 5.2) is The Verge's per-token arithmetic with the outlet's own token-count hedge — never present it as a total-cost fact or as Google's claim. Keep the availability gates visible: paying developers and enterprises via the API, Google AI Ultra subscribers in the consumer surfaces, /swarm a preview, and The Verge's lower-tier/free gap attributed to The Verge. 'Marking the first time that milestone has been reached' is Google's own framing of its own 0% result.
Usable claims
- Google's official post announces: "Today, we’re making Gemini 4 Carbon available to developers, enterprises, scientists, and consumers. Carbon is our strongest model with frontier performance in cybersecurity capabilities and long-horizon agentic workflows across complex knowledge work, software engineering, and scientific reasoning." Per the post, "Two weeks ago, we released Carbon to a set of cybersecurity defenders via our Fairwind Program." and "Following our recent launch of Gemini 4 Argon, Carbon pushes the frontier even further." — while "Argon remains the ideal choice for the majority of software engineering and knowledge work, while Carbon is built for tasks that test the limits of a model’s intelligence like complex, multi-step coding and debugging and open-ended scientific research and discovery", with Carbon doing that "using more tokens than Argon to explore thoroughly, reason deeply, and run extended verification loops". On benchmarks, per the post, "It ranks #1 on CursorBench 4.0, Perplexity’s WANDR, and Zapier’s AutomationBench at more than 2X lower cost per task compared to the next best external model."; and on the new Multimodal Knowledge Agent benchmark — per the post, Google "partnered with Artificial Analysis to launch the new Multimodal Knowledge Agent (MKA) benchmark that tests a model’s multimodal performance in agentic workflows" — "Carbon ranks #2 while being more than 6X lower in average cost per task compared to the top performing model." Per the post, Google is "previewing /swarm in Google Antigravity, turning a single prompt into a coordinated team of agents you can track in real time" — a preview, not a general release. Availability per the post: "Gemini 4 Carbon is available via the API in Google AI Studio and Gemini for business for all paying developers and enterprises" at the same introductory price as Gemini 4 Argon ($2 per million input tokens and $10 per million output tokens), and "It is also rolling out to all Google AI Ultra subscribers in Google Antigravity, the Gemini app, and AI Mode in Search." The post's own footnote states "After the introductory period expires on November 30, 2026, the price of $4 per 1M input tokens and $20 per 1M output tokens will apply." and "Applies to the Carbon model intended for public access only." The post also carries Google's own safety numbers: "Carbon blocked 99% of attack trajectories from our adaptive red-teaming agent that can iteratively revisit its attack plans after each block"; "The Carbon safeguards lowered the per-attempt breach rate to just 0.8% (down from 3.1% on Argon)."; "Our safeguards also lowered the automated cyberattack query success rate to 2.17% in Gray Swan’s testing, compared to 9.86% on Argon."; "And external testing by Trajectory Labs found zero new universal jailbreaks across 132 hours and over 13,000 requests."; that "on our 375-task agentic honesty benchmark, it passed 92% of tasks (compared to 70% for Argon)"; and "Two weeks ago, we shared that Carbon achieved a 0% prompt injection attack success rate on the Gray Swan IPI benchmark — marking the first time that milestone has been reached." — every figure a Google claim about its own model. The Verge relays the rollout — "Google is rolling out its latest AI model, Gemini 4 Carbon, to the general public after releasing it to a select group of “cybersecurity defenders” two weeks ago." — and adds its own context: the intro price is "around a fifth the cost of OpenAI’s GPT-6 Astra and a quarter the cost of Anthropic’s Fable 5.2 model, which launched yesterday", and "It is not clear when — or if — Carbon will be available to lower tier subscribers or free users."
Evidence pipeline
From the news
Breakdown
Two opened sources, one primary: Google's post carries every announcement fact, every benchmark and safety number, and the pricing footnote; The Verge is the relay adding the rival-cost arithmetic, the cherry-picking caution, and the lower-tier gap. What is a vendor claim: all the numbers — the #1/#2 rankings, the 'more than 2X' and 'more than 6X' cost multiples against unnamed referents, the safety rates, and the 'first time that milestone has been reached' framing. What The Verge adds: the GPT-6 Astra / Fable 5.2 comparison (per-token, hedged by the outlet itself on token counts), and the "It is not clear when — or if — Carbon will be available to lower tier subscribers or free users." gap. What the post never says: any comparison to a named competitor's model, any availability promise for lower tiers or free users, and any price beyond the footnote's November 30, 2026 step-up. The /swarm feature is a preview in the post's own words, and the 0% prompt-injection result is the post's restatement of what it shared two weeks ago.
Sources
Risks
- Before you publish, keep every benchmark and safety number inside its 'per Google' attribution with its 'more than' bound intact, and say plainly that no opened source independently benchmarked any of them. Keep the rival-cost comparison labeled as The Verge's per-token arithmetic with the outlet's token-count hedge attached. Check the availability wording line by line: paying developers and enterprises via the API, AI Ultra subscribers in the consumer surfaces, /swarm is a preview — and attribute the lower-tier/free gap to The Verge. Show the intro price with its expiry on screen: $2/$10 now, $4/$20 after November 30, 2026, per Google's footnote.
Demo ideas
- An Argon-versus-Carbon positioning card built on the post's own split — "Argon remains the ideal choice for the majority of software engineering and knowledge work" while Carbon "is built for tasks that test the limits of a model’s intelligence" and works "using more tokens than Argon to explore thoroughly, reason deeply, and run extended verification loops" — with the token-consumption trade-off spelled out as the cost angle the rankings don't capture.
- A benchmark-scorecard card where every row carries its attribution: #1 on CursorBench 4.0, Perplexity's WANDR, and Zapier's AutomationBench at "more than 2X lower cost per task", #2 on the MKA benchmark at "more than 6X lower" — with the unnamed referents on screen ('the next best external model', 'the top performing model') and The Verge's cherry-pick caution as the closing line.
- An availability-map card: the API (Google AI Studio and Gemini for business) for paying developers and enterprises; the Gemini app, AI Mode in Search, and Antigravity for Google AI Ultra subscribers; /swarm as a preview in Antigravity — with The Verge's "It is not clear when — or if — Carbon will be available to lower tier subscribers or free users." as the open question at the bottom.