Google released Gemini 4 Hydrogen, a model developed by Google DeepMind for routine agent loops, positioned below its two frontier models, Gemini 4 Carbon and Gemini 4 Argon. Its model ID is gemini-4-hydrogen. Reaching it through Google Antigravity requires a Google AI Pro or Ultra subscription.
Hydrogen finishes tasks 2.5 times faster than Gemini 3.8 Flash at its default high thinking level, according to Google.
Thinking tokens count as output, at $7.50 for each million tokens; input costs $1.50 per million, and neither price changes across the entire one-million-token window.
Google is already running a promotional ladder, with those rates landing inside it. Gemini 3.8 Flash charges $0.75 for input and $3.75 for output until 31 December 2026, while Gemini 4 Argon charges $2 and $10 until 30 November 2026.
| Model | Input per 1M tokens | Output per 1M tokens | Rate applies |
|---|---|---|---|
| Gemini 4 Hydrogen | $1.50 | $7.50 | Flat across the 1M-token window |
| Gemini 3.8 Flash | $0.75 | $3.75 | Until 31 December 2026 |
| Gemini 3.8 Flash from 1 January 2027 | $1.50 | $7.50 | Matches Hydrogen |
| Gemini 4 Argon | $2 | $10 | Until 30 November 2026 |
| Gemini 4 Argon from 30 November 2026 | $4 | $20 | Doubles |
Compared to 3.8 Flash, it finishes tasks 2.5x faster. Also has 4 thinking levels, so you only wait and pay for the reasoning depth a task requires
Four thinking levels ship with the model: low, medium, high and a new xhigh. High is the default, and a minimal setting now returns an error.
Inputs cover text, images, audio, video, PDFs, tables and charts, while output is text. One million tokens of context are shared by input, thinking and output together.
The model is aimed at research, data analysis and work involving documents, spreadsheets, charts, audio and video. Google said it can sit alongside other models handling coordination, coding, testing and verification.
Google's benchmarks put Hydrogen ahead of 3.8 Flash
Terminal-Bench 4.1 gave Hydrogen 57% against 18.2% for Gemini 3.8 Flash, Google said.
According to Google, fabricated claims fell by more than five times against 3.8 Flash, and the model scored 96.7% on an internal agent honesty evaluation.
In partner tests, Hydrogen ran between 2 and 2.8 times faster, Google said. The same tests recorded 12% to 23% fewer tokens.
Google said Hydrogen's score on the Artificial Analysis Intelligence Index was 51 at its default high thinking level, against 50 for Gemini 4 Argon. At the extra-high setting that figure rises to 54, Google said. Hindustan Times wrote that Claude Fable 5.2 leads the same leaderboard on 65, with Gemini 4 Carbon on 60.
Released alongside the model, benchmark tables put Hydrogen against Claude Sonnet 5.5 and GPT-6.1 Terra on three agentic tests. Hydrogen was ahead on each.
| Benchmark | Gemini 4 Hydrogen | Claude Sonnet 5.5 | GPT-6.1 Terra |
|---|---|---|---|
| Agents' Last Exam v1.1 | 71.6% | 60.4% | 69.8% |
| HLE-Diamond | 58.8% | 53.3% | 47.3% |
| SWE-sweep | 13.5% | 3.1% | 5.8% |
Orchestrator pairs Carbon with Hydrogen
Inside Google Antigravity, a new feature called Orchestrator hands a single task to more than one model. Carbon explores the problem, resolves ambiguity and writes the plan; Hydrogen edits files, updates packages and runs tests.
The Carbon-Hydrogen pairing reached near-Carbon success rates on Terminal-Bench while using 24% less quota and finishing 19% faster than Carbon alone, Google said.
While our frontier models, Gemini 4 Carbon and Gemini 4 Argon, excel at deep, long-horizon reasoning on your hardest problems, Hydrogen is built for rapid execution across diverse agentic workflows
Hydrogen is built to stand on its own as a model for everyday work — or to pair with Carbon and Argon as the high-efficiency execution engine in multi-model workflows
Google claims 2.3 times more resilience to indirect prompt injection on Gray Swan's benchmark. It also says the model is two times less likely to circumvent permission boundaries than 3.8 Flash, and reduces excessive actions in agentic tasks by up to 81%.
Gemini 4 Hydrogen brings improvements from our latest Gemini 4 safety and security stack to our broadest audience yet, combining built-in training with real-time surface safeguards to help mitigate both misalignment and misuse
Availability and migration
Free users of the Gemini app see Hydrogen only through automatic model selection, when Google judges it fits the prompt; AI Pro and Ultra subscribers can choose it manually. Inside Google Sheets, Gemini remains limited to those two paid tiers.
Moving a workload off 3.8 Flash means dropping the temperature, top_p and top_k settings, removing cached_content fields and raising hardcoded max_output_tokens caps to 65,536.
For prompts of at least 4,096 tokens, implicit caching lowers the input price to $0.075 per million tokens. Explicit caching isn't available.
Adapted from Gemini 4 Carbon, the model has a knowledge cutoff of March 2026, according to Google. The company doesn't disclose its parameter count, architecture or training hardware.
Hallucination remains a problem, as it does for generative AI agents generally. Google said work on jailbreak resistance is continuing.
Google Cloud documentation lists Hydrogen as generally available, in the category of a short-term release.
What we know
- Gemini 4 Hydrogen is priced at $1.50 per million input tokens and $7.50 per million output tokens, flat across the 1M-token window.
- Google lists the model as generally available in its Cloud documentation.
- Free Gemini app users reach Hydrogen only through automatic model selection; AI Pro and Ultra subscribers can pick it manually.
Still unclear
- Whether Gemini 4 Hydrogen will be offered on Vertex AI.
- Why Google Cloud documentation labels it a short-term release.
- Hydrogen's parameter count, architecture and training hardware, which Google does not disclose.
Gemini 4 Argon's promotional $2/$10 rate ends on 30 November 2026. It then doubles to $4/$20.
Gemini 3.8 Flash's $0.75/$3.75 rate ends on 31 December 2026. From 1 January 2027, it rises to Hydrogen's $1.50/$7.50.
- 30 November 2026Argon doubles
Gemini 4 Argon moves from $2/$10 to $4/$20 per million tokens.
- 31 December 2026Flash discount expires
Gemini 3.8 Flash's $0.75/$3.75 rate ends.
- 1 January 2027Flash matches Hydrogen
Gemini 3.8 Flash rises to $1.50/$7.50, the Hydrogen rate.
