Google released three Gemini 4 models in six weeks: Argon (generally available 13 October 2026), Carbon (available 10 November) and Hydrogen (12 November). None of them is simply called “Gemini 4”, which has caused confusion. The short answer: Carbon is the most capable, Argon is the general-purpose default and Hydrogen is the fastest and cheapest. The details below are based on Google’s announcements as summarised by SmartScope and other coverage; benchmark figures are Google-reported.
Gemini 4 at a glance
| Argon | Carbon | Hydrogen | |
|---|---|---|---|
| Available | 13 Oct 2026 | 10 Nov 2026 | 12 Nov 2026 |
| Best for | Most software development and knowledge work | Hardest multi-step coding, debugging, research and cybersecurity | Fast everyday coding and agent workflows |
| API price per 1M tokens (input / output) | $2 / $10 introductory, then $4 / $20 | $2 / $10 introductory, then $4 / $20 | $1.50 / $7.50 |
| Artificial Analysis Intelligence Index | 50 | 59 | 51 (54 at highest setting) |
| Context window | About 1 million tokens shared across input, thinking and output | ||
The introductory price for Argon and Carbon runs until 30 November 2026.
Gemini 4 Carbon: the strongest model
Google calls Carbon its strongest model. It is based on Argon and thinks longer, using more tokens per task. Google reports 65.2% on Terminal-Bench 4.0, 81.5% on OSWorld-2.1 for computer use and 88.2% on SEC-bench Pro for security work. Carbon does not win everything: Google’s own tables show GPT-6 Astra and Claude Opus 5.5 ahead on some measures. Access is the main limitation. In the Gemini app and the Antigravity coding platform, Carbon is only for Google AI Ultra subscribers; developers can use it on the paid Gemini API tier.
Gemini 4 Argon: the default for most people
Google describes Argon as “the ideal choice for most software development and knowledge work.” It is available to developers and enterprises through the API, and in the Gemini app for AI Pro and Ultra subscribers. At the same introductory price as Carbon, Argon makes sense when you want strong results with lower latency and fewer tokens per answer.
Gemini 4 Hydrogen: fast and cheap
Hydrogen is adapted from Carbon and tuned for speed and cost. Google reports 57% on Terminal-Bench 4.1, compared with 18.2% for Gemini 3.8 Flash, as Interesting Engineering reported. It is the most widely available model: all AI Studio tiers, the Gemini API, Antigravity for paid plans and the Gemini app model picker, with free users sometimes routed to it automatically.
Using Carbon and Hydrogen together
Google’s Antigravity platform now has an Orchestrator feature that switches between models within one session. The suggested pattern is Carbon as the coordinator that plans the work and Hydrogen as the worker and validator agents. Google says this came close to Carbon’s success rate on Terminal-Bench 4.0 while using 24% less quota and finishing 19% faster.
Which should you choose?
- Choose Hydrogen for high-volume coding help, agents that run many steps and anything cost-sensitive.
- Choose Argon for everyday writing, analysis and development if you have Pro.
- Choose Carbon for the hardest debugging, research or security tasks, if you have Ultra or paid API access.
Rival flagships launched in the same weeks, so compare before committing. Our Claude Fable 5.2 vs GPT-6.2 Astra comparison covers the other two top models.
FAQ
Is there a model called “Gemini 4”?
No. Google’s Gemini 4 family has three named models: Argon, Carbon and Hydrogen. The API does not accept a plain “gemini-4” identifier.
Which Gemini 4 model is free?
Hydrogen is the only one free users may get, through automatic routing in the Gemini app and the free AI Studio tier.
Sources
- Google DeepMind: Introducing Gemini 4 Hydrogen
- Google: Gemini 4 Carbon is now available
- SmartScope: Gemini 4 Hydrogen vs Argon vs Carbon
- Interesting Engineering: Gemini 4 Hydrogen benchmark
Published
Spotted an error? Read our corrections policy or email editorial@perplexityaimagazine.com.