Google released Gemini 4 Hydrogen on November 12, 2026. Google’s announcement, written by Koray Kavukcuoglu of Google DeepMind, pitches it as the fast execution model of the Gemini 4 family. Gemini 4 Carbon and Gemini 4 Argon handle deep, long-horizon reasoning, while Hydrogen writes code, runs tests and checks changes. The API model code is gemini-4-hydrogen.
Gemini 4 Hydrogen pricing is flat across the whole 1M-token context: $1.50 per million input tokens and $7.50 per million output tokens on the paid tier, with thinking tokens billed as output. Google says Hydrogen scores 51 on the Artificial Analysis Intelligence Index at its default high thinking level, one point above Argon’s 50, and 54 at the new xhigh level.
Hydrogen is also Google’s pick for teams moving off Gemini 3.8 Flash. That makes the price gap between the two models the first thing to check, because for the rest of 2026 Gemini 3.8 Flash costs half as much.
Gemini 4 Hydrogen at a glance
The Gemini API models page and pricing page list these details:
gemini-4-hydrogen
| Model code | gemini-4-hydrogen |
|---|---|
| Context window | 1M tokens |
| Max output | 65K tokens, with long-decode output up to 1M tokens |
| Inputs | Text, image, audio, video, PDF |
| Thinking levels | low, medium, high (default), xhigh |
| Paid tier, Standard | $1.50 input / $7.50 output per 1M tokens, flat across 1M context |
| Context caching | $0.075 per 1M tokens, implicit caching only |
| Free tier | Yes, in the Gemini API (free tier data is used to improve Google products) |
Google’s pricing page lists five ways to pay for the same model:
| Tier | Input per 1M | Output per 1M (incl. thinking) | Cached input per 1M |
|---|---|---|---|
| Standard (paid) | $1.50 | $7.50 | $0.075 |
| Batch | $0.75 | $3.75 | $0.0375 |
| Flex | $0.75 | $3.75 | $0.0375 |
| Priority | $2.70 | $13.50 | $0.135 |
| Free tier | Free | Free | Free |
The pricing page adds a note that applies to every Hydrogen bill. Every token the model processes is counted, including thinking tokens. With grounding, that also covers internal tool system instructions and function declarations. Grounding with Google Search gets 5,000 free requests per month, shared across all Gemini 3 and newer models, then costs $14 per 1,000 requests, with a daily cap of 1,500 requests. Explicit context caching isn’t supported. Caching happens implicitly, so you can’t pin a long system prompt or codebase dump the way explicit caching allows.
Where Hydrogen sits against Carbon and 3.8 Flash
Price is the part of the launch that changes most over the next seven weeks, because the two models on either side of Hydrogen have dated prices:
| Model | Input per 1M | Output per 1M | Price change |
|---|---|---|---|
| Gemini 4 Carbon | $2.00 | $10.00 | Rises to $4.00 / $20.00 on December 1, 2026 |
| Gemini 4 Hydrogen | $1.50 | $7.50 | None listed |
| Gemini 3.8 Flash | $0.75 | $3.75 | Rises to $1.50 / $7.50 on January 1, 2027 |
From January 1, 2027, Gemini 3.8 Flash and Gemini 4 Hydrogen carry the same list price per token. Until then, Hydrogen costs twice as much per token as 3.8 Flash. Carbon, which we covered in our Gemini 4 Carbon launch post, sits only a third above Hydrogen until November 30 and then doubles.
Here is plain arithmetic for one agent task that reads 2M input tokens and writes 200K output tokens, at Standard paid-tier list prices. It assumes each model uses the same number of tokens, which real runs won’t do:
- Gemini 4 Hydrogen: 2 × $1.50 + 0.2 × $7.50 = $4.50
- Gemini 3.8 Flash (through December 31): 2 × $0.75 + 0.2 × $3.75 = $2.25
- Gemini 4 Carbon (through November 30): 2 × $2.00 + 0.2 × $10.00 = $6.00
- Gemini 4 Carbon (from December 1): 2 × $4.00 + 0.2 × $20.00 = $12.00
Token counts move the result. Google says partners running Hydrogen on graph, SQL and data-heavy enterprise workflows saw runs finish 2x to 2.8x faster with 12% to 23% fewer tokens than 3.8 Flash. A task that burns fewer tokens on Hydrogen narrows the per-token gap, but that’s a partner result, not a figure you can bill against.
What Google reports for coding and agents
All of these figures come from Google’s announcement, and some come from Google’s own or partner evaluations. They are vendor claims until independent runs confirm them.
Against Gemini 3.8 Flash, Google says Hydrogen finishes tasks 2.5 times faster at its default high setting, raises Terminal-Bench 4.1 from 18.2% to 57%, and cuts fabricated claims by more than 5x. Google also says the model batches file-read and search calls in parallel when it explores a large repository, instead of opening files one by one, which is where the partner speed-ups above came from.
For agent and knowledge work, Google lists these results:
- HLE-Diamond: 58.8%
- Agent Arena (arena.ai’s live benchmark): ranked #5
- Vals Tax Agent Bench: 52.3%, #1 overall
- Vals Legal Research Bench: 55.8%, tied for #1
- Multimodal Knowledge Agent (Google’s new benchmark): 53.4%
- Agents’ Last Exam v1.1: 71.6%
Google’s safety section compares Hydrogen with 3.8 Flash too. Google says Hydrogen reduces overreach in agentic tasks by up to 81% and under-service by up to 79%. It cuts misaligned actions and task shortcutting across tool-use and coding evaluations by 12x. On impossible or blocked tasks, it is up to 2x less likely to bypass permission boundaries, and it reaches a 96.7% honesty rate on Google’s internal agent honesty evaluation. On the Gray Swan indirect prompt injection benchmark, Google reports Hydrogen as more than 2.3x more resilient. For a model meant to run as an unattended worker agent, the permission-boundary and prompt-injection numbers are the ones to test on your own setup.
Hydrogen as the worker under a planner
Google built the launch around multi-model sessions. Its announcement describes a frontier model (Carbon or Argon) making the design calls, and Hydrogen doing the fast loop of editing code, running tests and verifying changes.
Google Antigravity gets a new Orchestrator feature for this. Orchestrator switches between models such as Carbon and Hydrogen inside one session. On Terminal-Bench 4.0, Google says Orchestrator reached near-Carbon task success while using 24% less quota on average and finishing 19% faster. For parallel agent work, Google describes Carbon as the coordinator and Hydrogen as the worker and verifier agents that build and test changes across isolated workspaces.
The same split works outside Antigravity if your agent tool lets you set a separate model for subagents or verification steps. The planner keeps the expensive Carbon tokens to a few calls, and the many small edit-and-test calls go to gemini-4-hydrogen. We looked at another cheap worker-tier option in our Meta Muse Ember pricing post.
Where you can use it
Google says Hydrogen is rolling out globally from November 12:
Gemini 4 Hydrogen availability
- Gemini API, in Google AI Studio and Android Studio, on the free and paid tiers
- Google Antigravity, for paid Google AI Pro and Ultra subscribers
- Gemini for business, through the API or the model drop-down, plus Antigravity and Gemini in Google Sheets for eligible business subscriptions
- The Gemini app model drop-down; free users get Hydrogen through auto model selection when it fits the prompt
- Google Search for all users, and Gemini in Google Sheets starting with Google AI Pro and Ultra
The Gemini API models page also points new projects at Hydrogen. In its note on the older 2.5 Flash models, Google says that for new projects you should use Gemini 4 Hydrogen or 3.5 Flash-Lite. The same page says requests to the deprecated gemini-2.5-flash will automatically route to gemini-3.8-flash from December 1, 2026. If you still call 2.5 Flash, that date matters more than Hydrogen’s launch. For another recent Gemini API launch, see our Gemini Decisions API pricing notes.
Should you switch to Gemini 4 Hydrogen now?
If your agents run on Gemini 3.8 Flash and the bill matters more than speed, there’s no rush before January. Gemini 3.8 Flash is half Hydrogen’s price per token until December 31, 2026, and the two carry the same list price from January 1, 2027. Use the next seven weeks to run Hydrogen on your own tasks and measure tokens per task, test pass rates and wall-clock time against 3.8 Flash.
If you pay for Carbon on the edit-and-test loop, Hydrogen is worth testing sooner. Carbon’s price doubles on December 1, and Google’s own pitch is that Hydrogen takes the worker and verifier roles while Carbon keeps the planning. In both cases, set the thinking level on purpose, since the high default and the xhigh tier both bill thinking as output tokens.
