Gemini 4 Hydrogen vs Gemini 4 Carbon: Price, Speed, and Which One Fits Your Job
Gemini 4 Hydrogen costs $1.50 per million input tokens and $7.50 per million output tokens, flat. Gemini 4 Carbon costs $2 per million input and $10 per million output today — but only until 30 November 2026, after which it rises to $4 and $20. Both models sit in the 1-million-token context class, both carry the Gemini 4 name, and both were announced within two weeks of each other, yet Google positions them for different jobs: Hydrogen for rapid execution across everyday agentic work, Carbon as its strongest model for deep, long-horizon reasoning and cybersecurity. There is no winner here — as in our flagship AI model comparison — because no independent benchmarks exist for either model, and the numbers Google publishes for the two models measure different things.
The Dates, Corrected
Gemini 4 Hydrogen was introduced on 12 November 2026, in Google's blog post "Introducing Gemini 4 Hydrogen for Agentic Workflows," dated 12 November 2026. Gemini 4 Carbon was announced on 28 October 2026, in Google's blog post "Introducing Gemini 4 Carbon," dated 28 October 2026. The date 29 October 2026, which appears in some coverage, is wrong by one day: 29 October is when press coverage was published, not when Google announced Carbon. Carbon then reached general availability on 10 November 2026, when Google published "Gemini 4 Carbon is now available."
Price: The Gap Today, and the Gap From 1 December
All prices below come from Google's own Gemini API pricing page. On the standard paid tier, Hydrogen charges $1.50 per 1 million input tokens and $7.50 per 1 million output tokens, flat. Cached input costs $0.075 per million tokens, using implicit caching only. Batch and Flex is $0.75 input and $3.75 output, and Priority is $2.70 and $13.50. No expiry is listed for Hydrogen, and its free tier is free of charge for input, output, and cached input on the API. Carbon is higher, and it changes: $2.00 input and $10.00 output through 30 November 2026, then $4.00 and $20.00 from 1 December 2026. Cached input is $0.10, then $0.20; Batch and Flex is $1.00 and $5.00, then $2.00 and $10.00; Priority is $3.60 and $18.00, then $7.20 and $36.00. Carbon has no free tier: Google states that "Gemini 4 Carbon is not available on the Free tier." Today Carbon costs 33 percent more than Hydrogen on input and output; from 1 December that gap doubles.
![]() |
| Two smartphones side by side — Gemini 4 Hydrogen and Gemini 4 Carbon compared. |
| Gemini 4 Hydrogen | Gemini 4 Carbon | |
|---|---|---|
| Standard paid tier, input / output per 1M tokens | $1.50 / $7.50, flat | $2.00 / $10.00 through 30 Nov 2026; $4.00 / $20.00 from 1 Dec 2026 |
| Cached input per 1M tokens | $0.075 | $0.10 through 30 Nov; $0.20 from 1 Dec |
| Batch / Flex per 1M tokens | $0.75 / $3.75 | $1.00 / $5.00 through 30 Nov; $2.00 / $10.00 from 1 Dec |
| Priority per 1M tokens | $2.70 / $13.50 | $3.60 / $18.00 through 30 Nov; $7.20 / $36.00 from 1 Dec |
| Free tier | Free of charge on the API | Not available on the Free tier |
| Context | 1M-token class; native 1M input, up to 1M output | 1M context; long-decode generation up to 1M tokens |
| Model ID | gemini-4-hydrogen | gemini-4-carbon |
Availability: Who Can Use Each One, and Where
Hydrogen is rolling out globally from 12 November across AI Studio, the Gemini API, Android Studio, Gemini for business, the Gemini app, and Google Search for all users, plus Google Antigravity and Google Sheets for paid Google AI Pro and Ultra subscribers. A free tier exists on the API. Carbon's rollout has been slower and more gated. At announcement on 28 October, it was available only to a subset of cyber defenders through the Fairwind Program, with a phased rollout to trusted partners. From 10 November, it is available via the API in AI Studio and Gemini for business for all paying developers and enterprises, rolling out to all Google AI Ultra subscribers in Antigravity, the Gemini app, and AI Mode in Search. The cyber version remains via Fairwind. Hydrogen is broadly available, including free, while Carbon is paid-tier only on the API and needs Ultra for the consumer app.
Benchmarks: Google's Claims, Not Independent Tests
Every benchmark figure here is Google's own claim, not an independent test. No independent benchmarks exist for either model, so no verdict can be drawn from these numbers alone. For Hydrogen, Google reports an Artificial Analysis Intelligence Index score of 51 at default high thinking, against 50 for Argon, and 54 at the new xhigh level. On Terminal-Bench 4.1, Hydrogen scores 57 percent, against 18.2 percent for Gemini 3.8 Flash, and Google says it is 2.5 times faster than 3.8 Flash at high thinking. Other Google-reported Hydrogen figures are 58.8 percent on HLE-Diamond; 52.3 percent on the Vals Tax Agent Bench, ranked first; 55.8 percent on the Vals Legal Research Bench, tied for first For Carbon, Google reports an Artificial Analysis Intelligence Index v5.0 score of 59, ranked first, plus 88.2 percent on SEC-bench Pro, 54.8 percent on ExploitGym, and 71.0 percent on CWE-bench. On WebDev Arena it ranks fourth at 1768, up 89 Elo, and Google reports it first on CursorBench 4.0, Perplexity WANDR, and Zapier AutomationBench. Two warnings apply: Google's own note says the cyber benchmarks were evaluated without guardrails, and the head-to-head comparison does not work — Hydrogen at 51 and Carbon at 59 use different thinking settings and benchmark sets, so no single winner can be derived from them.
![]() |
| A developer comparing two laptops — choosing between Hydrogen and Carbon comes down to the job, not the name. |
What Each Model Is Built For
Carbon is described as Google's strongest model, built for deep, long-horizon reasoning, the hardest multi-step coding and debugging, scientific research, and cybersecurity across the full defence lifecycle — find, then validate and exploit, then patch. Google says Carbon "uses more tokens than Argon to explore thoroughly, reason deeply, and run extended verification loops," and reports internal teams running it "for days" on complex engineering tasks. Hydrogen is built for the opposite shape of work: "rapid execution across diverse agentic workflows" — everyday coding, debugging, research, data analysis, and web development. Google calls it a "major standalone upgrade" and the recommended successor to Gemini 3.8 Flash. It offers four thinking levels — low, medium, high as the default, and a new xhigh tier — which Carbon and Argon do not offer; xhigh lifts its Artificial Analysis score from 51 to 54. Google also positions the two as a pair: Carbon as coordinator and Hydrogen as worker and verifier.
Context, Thinking, and the Practical Choice
Both models are in the 1-million-token context class. Hydrogen has a native 1-million-token input with up to 1 million output tokens, and Google's pricing page calls it a unified 1-million-token context window. Carbon has a 1-million-token context with long-decode generation up to 1 million tokens. The context difference is not the deciding factor — the price, the free tier, and the intended job are. For everyday coding, debugging, research, data analysis, or web development, especially where cost matters, Hydrogen is the model Google built for that job, at a flat $1.50 and $7.50 with a free API tier. For the hardest multi-step coding, long-horizon research, or cybersecurity, Carbon is the model Google built for that job, if the budget can carry a paid tier that doubles on 1 December. Hydrogen can be tried free on the API; Carbon cannot be used free at all.
![]() |
| A Google Pixel phone showing the Gemini app — Hydrogen is broadly available, including free, while Carbon is paid-tier only. |
Key Takeaways
- Hydrogen costs $1.50 per million input tokens and $7.50 per million output tokens, flat, with a free tier on the API. Carbon costs $2 and $10 through 30 November 2026, then $4 and $20 from 1 December, with no free tier.
- Carbon was announced on 28 October 2026, not 29 October, and reached general availability on 10 November. Hydrogen launched on 12 November 2026.
- Both models have a 1-million-token context, but Google positions them differently: Hydrogen for rapid execution, Carbon as its strongest model for deep reasoning and cybersecurity.
- All benchmark figures are Google's own claims, not independent tests, and they measure different things — no winner can be derived from them. Carbon's cyber benchmarks were evaluated without guardrails.
- Hydrogen offers four thinking levels, including a new xhigh tier, and a free API tier. Carbon is paid-tier only and needs Ultra for the consumer app.
Conclusion
Hydrogen and Carbon are not rivals for one job. They are a fast execution engine and a deep reasoning engine, priced accordingly, with Carbon's introductory price expiring on 30 November, doubling on 1 December. The approach is to match the model to the job, watch the December price change, and wait for independent benchmarks before calling either one the better model. Until those tests exist, Google's own numbers are a description of what each model was built to do, not a ranking. The same caution applies to the GPT-6 Astra benchmark scores and GPT-6.1 Terra.



Comments
Post a Comment