API Market / AI News / Gemini 4 Hydrogen: what Google launched on November 12

Gemini 4 Hydrogen: what Google launched on November 12

Published: November 13, 2026 · Official

On November 12, 2026, Google DeepMind introduced Gemini 4 Hydrogen, a model for everyday coding and agentic workflows that sits alongside the frontier Gemini 4 Carbon and Gemini 4 Argon. Google says Hydrogen finishes tasks 2.5 times faster than Gemini 3.8 Flash at the default thinking level, at a flat $1.50 per million input tokens and $7.50 per million output tokens. It is rolling out now in the Gemini API, the Gemini app, Google Search and Google Antigravity.

What the model is for

The announcement came from Koray Kavukcuoglu, SVP of Google DeepMind and Chief AI Architect at Google, who framed Hydrogen as bringing cutting-edge intelligence to everyone, with a focus on speed and efficiency for everyday coding and agentic work.

The positioning is explicit: Gemini 4 Carbon and Gemini 4 Argon handle deep, long-horizon reasoning on the hardest problems, while Hydrogen is built for rapid execution — writing code, running tests, verifying changes, doing research and driving tool-using agents.

According to Unite.AI, Hydrogen is adapted from Gemini 4 Carbon and accepts text, images, audio and video, with a March 2026 knowledge cutoff that may be less current in some domains. Google describes it as either a standalone model for everyday work or a high-efficiency execution engine paired with Carbon and Argon, which keep the architectural judgment and the subtle bug hunting.

Speed and the numbers Google reports

The headline comparison is with the previous Flash generation. Google says Hydrogen finishes tasks 2.5 times faster than Gemini 3.8 Flash at the default high setting, scores 57 percent on Terminal-Bench 4.1 against 18.2 percent, and cuts fabricated claims by more than fivefold. It also recovers better when a tool call fails or a search comes back empty.

On the Artificial Analysis Intelligence Index it scores 51 at high versus 50 for Argon, and 54 at xhigh. The Decoder notes that at its maximum thinking level Hydrogen trails Claude Sonnet 5.5 and GPT-6.1 Sol, which each score 55.

Other figures published by Google and rounded up by The Decoder:

One caveat comes from the model card: in the cheating-adjusted Terminal-Bench 4.1 table, Claude Sonnet 5.5 is listed at 63.9 percent, above Hydrogen at 57.0 percent. The jump over Google's own Flash model is steep, but Hydrogen does not lead every coding comparison. Partners testing early builds on graph, SQL and data-heavy enterprise workflows reported running 2x to 2.8x faster with 12 to 23 percent fewer tokens than Gemini 3.8 Flash.

Four thinking levels, pricing and API details

Hydrogen offers four thinking levels: low, medium, high by default, and a new xhigh. Google recommends high for most complex tasks and reserves xhigh for harder multistep problems where high falls short, noting that xhigh can overthink simple prompts while adding latency. The API rejects the minimal level; Argon offers only medium and high.

Pricing is flat across the full one-million-token window: $1.50 per million input tokens and $7.50 per million output tokens. Cached input costs $0.075 per million tokens through implicit caching, which applies automatically to prompts of at least 4,096 tokens; explicit context caching is not supported. Google AI Studio has a rate-limited free tier, and grounding with Google Search is billed per query.

Input, thinking and output share a single 1,048,576-token window. A synchronous request caps output at 65,536 tokens; longer output, up to a million tokens, requires background execution or client-side continuation. The API model ID is gemini-4-hydrogen. Developers should note that the model ignores temperature, top_p and top_k, and tool calls must be declared natively, otherwise a MALFORMED_FUNCTION_CALL error is possible.

Orchestrator: Carbon plans, Hydrogen executes

Alongside the model, Google introduced Orchestrator in Google Antigravity, which switches between models within a single session instead of committing a whole task to one model. Carbon can explore dependencies and produce a plan in a read-only phase, Hydrogen edits code, updates packages and runs tests, and control can return to Carbon when execution hits friction.

On Terminal-Bench 4.0, Google says this reaches near-Carbon task success while consuming 24 percent less quota and finishing 19 percent faster on average. For parallel work, Carbon coordinates while Hydrogen powers worker and verifier agents in isolated workspaces. Orchestrator initially rolls out to AI Ultra subscribers, with enterprise access to both through Gemini for business.

Where it is available, and the fine print

The rollout is global: the Gemini API on all tiers, Google AI Studio, Android Studio on the paid tier, Google Antigravity for paid Google AI Pro and Ultra subscribers, Gemini for business, Gemini in Google Sheets with an eligible subscription, the Gemini app and Google Search. In the Gemini app, free users get Hydrogen through automatic model selection when Google considers it appropriate.

In Search, Google calls it the biggest model upgrade yet: better handling of complex multipart questions and typos, an upgraded query fan-out technique that surfaces a wider range of links, and more visual answers with tables, images and custom tools, across all languages supported by Google's AI models.

The model card lists known limitations: hallucinations, occasional slowness or timeouts, and higher token use at elevated thinking levels. Automated comparisons also record a 5.57 percentage point increase in unjustified refusals versus Gemini 3.8 Flash. On safety, Google reports cuts of up to 81 percent in agentic overreach and 79 percent in under-service, more than 2.3x resilience to indirect prompt injection on the Gray Swan benchmark, and being up to 2x less likely to bypass permission boundaries. The frontier-safety assessment found no material general-capability increase over Argon; Hydrogen stayed below the critical capability levels for biological and cybersecurity risk but reached the alert thresholds in both. Fuller safety documentation is promised later.

What to use right now

You do not have to wait for the rollout: similar work can be done today with models already in the API Market catalog. Gemini 4 Carbon is Google's senior model with long context, suited to heavy reasoning and large documents. Gemini 3.8 Flash is the very fast Google option for high-volume jobs — the model Hydrogen is meant to replace. Claude Sonnet 5.5 is worth keeping for code and text work: in Google's own comparison, Sonnet scores higher on terminal tasks and computer use.

In API Market all of these are reachable with a single key, paid in rubles and without a VPN. New Google models appear in the catalog as access opens, so keep an eye on the API Market model catalog.

Models you can use right now

ModelAPI Market price
Gemini 4 Carbon
gemini-4-carbon		input $0.40, output $2.00 per 1M tokens
Gemini 3.8 Flash
gemini-3.8-flash		input $0.07, output $0.35 per 1M tokens
Claude Sonnet 5.5
claude-sonnet-5-5		input $0.45, output $2.25 per 1M tokens

Try these models on API Market

One API key for GPT, Claude, Gemini, DeepSeek, Grok and more. OpenAI-compatible endpoint, pay as you go, top-ups from about $0.15.

Get an API keyAll models

Connect in a minute

from openai import OpenAI

client = OpenAI(
    base_url="https://api-market1.shop/v1",
    api_key="sk-your-key",
)

reply = client.chat.completions.create(
    model="gemini-4-carbon",
    messages=[{"role": "user", "content": "Hi! Tell me about yourself."}],
)
print(reply.choices[0].message.content)

FAQ

What is Gemini 4 Hydrogen?

A new Google model announced on November 12, 2026. It targets fast everyday coding and agentic workflows and complements the frontier Gemini 4 Carbon and Gemini 4 Argon.

How much does Gemini 4 Hydrogen cost?

$1.50 per million input tokens and $7.50 per million output tokens, flat across the full one-million-token context window. Cached input costs $0.075 per million tokens.

How is Hydrogen different from Carbon and Argon?

Google positions Carbon and Argon for deep, long-horizon reasoning and Hydrogen as the fast execution model for writing code, running tests, verifying changes and using tools. Hydrogen is adapted from Carbon.

Which thinking levels does it offer?

Four: low, medium, high by default, and a new xhigh. Google recommends high for most complex tasks and xhigh for harder multistep problems where high falls short.

How much faster is it than Gemini 3.8 Flash?

Google says it finishes tasks 2.5 times faster at the default high setting and scores 57 percent on Terminal-Bench 4.1 versus 18.2 percent for Flash.

Sources

More news

Русская версия