Gemini 4 Hydrogen API

Gemini 4 Hydrogen at a glance

Release

Released and available on 2026-11-12: every tier of the Gemini API and AI Studio (free tier included), Android Studio and Gemini for business; the Gemini app (free users reach it through auto model selection); Search for all users; Antigravity, Gemini in Chrome and Gemini in Google Sheets for Google AI Pro and Ultra subscribers

Positioning

Fast and efficient, for everyday coding and agentic workflows, and Google’s recommended upgrade from Gemini 3.8 Flash and earlier models. Google says it finishes tasks 2.5x faster than 3.8 Flash at the default high level and cuts fabricated claims by more than 5x; the suggested pairing is Carbon to plan and Hydrogen to edit code and run tests

Context and output

1,048,576-token context window shared by input, thinking and output; up to 65,536 output tokens per synchronous request, about 1M with long decode (background mode). Accepts text, images, audio, video and PDF and outputs text; knowledge cutoff March 2026

Thinking levels

Four levels, low / medium / high / xhigh, with high as the API default; xhigh is new, and Google says to reserve it for hard multi-step problems where high falls short. minimal is no longer accepted and returns an error. Thinking tokens are billed at the output rate, and max_output_tokens caps thinking and output together

Official pricing

$1.50 input / $7.50 output per million tokens, $0.075 cache read; no introductory price and no long-context surcharge. Batch and Flex are half price and Priority is 1.8x; the API has a rate-limited free tier. Implicit caching only (prompts of at least 4,096 tokens). For comparison, 3.8 Flash is $0.75 / $3.75 through 2026-12-31 and $1.50 / $7.50 from 2027-01-01

API features

Function calling, structured outputs, code execution, Google Search and Maps grounding, URL context, file search, computer use, Batch, Flex and Priority; no Live API, image or audio generation, or explicit caching. temperature, top_p and top_k are ignored

Subscriptions

Antigravity’s model table lists Hydrogen on every plan, including free and AI Plus, while the launch post names paid AI Pro and Ultra and Antigravity’s lead says access will expand over time. Argon needs Pro; Carbon and the new Orchestrator (Carbon plans, Hydrogen executes) need Ultra

Model ID

gemini-4-hydrogen (the only stable version so far)

Checked 2026-11-13 · Sources: Model card · Gemini API · Model · Gemini API · What’s new · Gemini API · Pricing · Google · Antigravity · Artificial Analysis

Official benchmarks: Gemini 4 Hydrogen vs Claude Sonnet 5.5 vs GPT-6.1 Terra

From Google DeepMind’s Gemini 4 Hydrogen model card (as of November 2026). These are vendor-run results, Google also supplied the competitor scores, and the column headers do not state each model’s thinking level; “—” means Google listed no score.

Google’s table has 10 rows and Hydrogen leads 7 of them. Claude Sonnet 5.5 is higher on Terminal-Bench 4.1 (63.9% vs 57.0%), OSWorld-2.1 (83.9% vs 77.6%) and Terminal-Bench-Science (59.9% vs 51.4%), where GPT-6.1 Terra (58.3%) also beats Hydrogen. The row not shown above is the Vending-Bench 2 final balance: $14,549 for Hydrogen, $11,165 for Sonnet 5.5 and $9,107 for Terra. The model card has no Gemini 3.8 Flash column; Google’s launch post gives one figure, Terminal-Bench 4.1 at 57% for Hydrogen against 18.2% for 3.8 Flash.

Which to use: Gemini 4 Hydrogen, 3.8 Flash, Argon, Claude Sonnet 5.5 or GPT-6.1 Terra

Versus Gemini 3.8 Flash: should you upgrade

Hydrogen is the model that replaces 3.8 Flash. On price it costs twice as much today ($1.50 / $7.50 vs $0.75 / $3.75); from 2027-01-01 3.8 Flash rises to the same rates and its cache reads become the pricier ones ($0.15 vs $0.075). On capability, Google says it finishes tasks 2.5x faster at the default level and lifts Terminal-Bench 4.1 from 18.2% to 57%; Cursor measured CursorBench going from 39.6% to 54.3%; Box saw 5 points more accuracy on its enterprise document eval in under half the time with about 12% fewer tokens. To migrate, change the model ID and check four things: the default level is high rather than medium, minimal now errors, explicit caching is gone, and input and output share one 1M window.

Versus Gemini 4 Argon and Carbon

On the Artificial Analysis Intelligence Index (v5.0.1) Hydrogen scores 51 at its default high level, above Argon’s 50, and 54 at xhigh; at high it costs about 19% of Argon per task. The list price is lower too: Argon and Carbon are $2 / $10 through 2026-11-30 and $4 / $20 from 12-01. Carbon remains the strongest of the three at 60. Google’s split is Carbon and Argon for deep, long-horizon reasoning and Hydrogen for fast execution; Antigravity’s Orchestrator has Carbon plan and Hydrogen edit code and run tests, which Google says reaches near-Carbon success rates with 24% less quota and 19% less time. All three take the same request format, so switching only changes the ID; Argon and Carbon default to medium and stop at high.

Versus Claude Sonnet 5.5 and GPT-6.1 Terra

These are the two models Google compared Hydrogen with at launch. List prices are $2 / $10 for Sonnet 5.5, $1.50 / $7.50 for Hydrogen and $1 / $5 for GPT-6.1 Terra. On the Artificial Analysis Intelligence Index Hydrogen scores 54 at xhigh, 1 point behind Sonnet 5.5 (max), at about 32% of its cost per task; its AA-Omniscience hallucination rate is 8% against 43% for Sonnet 5.5. Sonnet 5.5 is stronger at agentic terminal work (64% vs 54% on Terminal-Bench 4.0) and leads on computer use. Against Terra, Hydrogen is higher on every row of Google’s table except Terminal-Bench-Science, but Terra’s list price is a third lower. Pick Sonnet 5.5 for terminal coding; Hydrogen is the better value for multimodal work, long documents and low hallucination.

Which thinking level, and is xhigh worth it

Google recommends the default high for most tasks and xhigh only for hard multi-step problems where high falls short; on simpler prompts xhigh adds latency and tokens and can overthink. Artificial Analysis measured all four levels: low scores 27 at $0.29 per task, medium 41 at $0.80, high 51 at $1.87 and xhigh 54 at $3.16. xhigh nearly doubles output tokens for 3 more points, makes no difference on Terminal-Bench 4.0 (54% at both) and lifts Terminal-Bench-Science from 41% to 50%. Use high for everyday coding, drop to medium for interactive work, and try xhigh on hard research-style problems. Through Wokey, choose the level with reasoning_effort; it runs at medium when you omit it, so pass high explicitly to match the Google API default.

Official migration notes

Independent evaluations and official figures

• Artificial Analysis: On Intelligence Index v5.0.1 Hydrogen scores 54 at xhigh, 16th of 78 models, against a median of 41; at the default high level it scores 51. Running the full index at xhigh took 304M output tokens (median 72.8M), which puts it among the most verbose models, at $3.16 and 13.5 minutes per task on average; at high it costs $1.87 per task. Output speed is 190 tokens per second, with 37.77 seconds to the first answer token at xhigh. Its 8% hallucination rate on AA-Omniscience is one of the lowest recorded. Artificial Analysis sums it up as “amongst the leading models in intelligence and reasonably priced”, but “slower than average per task and very verbose”. Every Wokey meter is 20% of Google’s list price, so the same task comes to about $0.37 at high and $0.63 at xhigh.
• Google launch post: Google says Hydrogen finishes tasks 2.5x faster than 3.8 Flash at its default high level, scores 57% on Terminal-Bench 4.1 (3.8 Flash 18.2%) and cuts fabricated claims by more than 5x; partners saw it run 2x to 2.8x faster than 3.8 Flash with 12% to 23% fewer tokens. It ranks 5th in Agent Arena, first on Vals Tax Agent Bench (52.3%) and tied for first on Vals Legal Research Bench (55.8%), and reaches a 96.7% honesty rate on Google’s internal agent evaluation. These are Google’s own figures.
• Cursor: Cursor added Hydrogen on launch day and says that at its default effort it scores 54.3% on CursorBench, up from 39.6% for Gemini 3.8 Flash.
• Box: Box CEO Aaron Levie says that on the Box AI Complex Work Eval (enterprise document tasks across finance, consulting, life sciences and other industries) Hydrogen is 5 percentage points more accurate than Gemini 3.8 Flash, finishes in less than half the time and uses about 12% fewer tokens on a typical task.

Calling Gemini 4 Hydrogen through Wokey: parameters and limits

This is how the Wokey gateway handles gemini-4-hydrogen. Coming from 3.8 Flash, Argon or Carbon, check the thinking level and max_tokens besides the model ID.

• The interface is OpenAI-compatible /v1/chat/completions: set the base URL to https://api.wokey.ai/v1, use your Wokey API key, and set model to gemini-4-hydrogen. opencode, OpenClaw, Hermes Agent and other OpenAI-compatible clients are configured the same way.
• reasoning_effort values low, medium, high and xhigh map directly to Hydrogen’s four thinking levels; without it, or with any other value such as minimal or max, the request runs at medium. Google’s own API defaults to high, so pass "high" explicitly to match it.
• Without max_tokens the limit defaults to 65,536, which is also Wokey’s current output ceiling; long decode and background mode are not supported yet. Thinking tokens count toward max_tokens and are billed at the output rate, so do not carry over a hard-coded low cap such as 8,192; to spend fewer tokens, lower the thinking level instead.
• xhigh thinks for a long time: Artificial Analysis measured about 38 seconds to the first answer token. Use stream: true and give the client a timeout of several minutes or more.
• Google ignores temperature, top_p and top_k, so you can leave them out.
• Text and image input are accepted today; audio, video and PDF input are not yet open on Wokey.
• Tool calling: declare tools in the tools field instead of prompting the model to print JSON tool calls, which Google says can return MALFORMED_FUNCTION_CALL. The gateway stores the thought signatures Hydrogen returns, so send back tool_calls and tool results as usual on the next turn.
• Structured outputs: use response_format with json_schema, and keep the schema within the OpenAPI subset Gemini accepts (type, properties, required, items, enum, description, and nullable via anyOf with null). json_object, or a schema with $ref, oneOf or pattern, currently returns 400 structured_output_supply_unavailable.
• Caching is Google’s implicit caching, with no cache_control; keep long sessions on a stable prefix to hit it. Cache reads cost 5% of the input rate.

Calling Gemini 4 Hydrogen through Wokey (/v1/chat/completions)

curl https://api.wokey.ai/v1/chat/completions \
  -H "Authorization: Bearer $WOKEY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gemini-4-hydrogen",
    "reasoning_effort": "high",
    "stream": true,
    "max_tokens": 65536,
    "messages": [{ "role": "user", "content": "Find the failing test in this repo and fix it" }]
  }'

Catalog pricing snapshot

Prices are per 1M tokens; request-time pricing controls settlement.

Official price source: https://aistudio.google.com/docs/pricing

Model capabilities

Context

1,048,576 tokens

Max output

65,536 tokens

Streaming

Supported

Tools

Supported

Vision

Supported

Supported API forms

/v1/chat/completions

Send a request

/v1/chat/completions

curl https://api.wokey.ai/v1/chat/completions \
  -H "Authorization: Bearer $WOKEY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gemini-4-hydrogen","messages":[{"role":"user","content":"Hello"}]}'

Use Gemini 4 Hydrogen in your tools

The gateway is https://api.wokey.ai (OpenAI-compatible clients usually take https://api.wokey.ai/v1), and the model ID is gemini-4-hydrogen. Open the guide for the client you use.

• Use Gemini 4 Hydrogen in opencode: OpenAI Chat Completions · base URL and API key setup
• Use Gemini 4 Hydrogen in DeepSeek Harness: OpenAI Chat Completions · base URL and API key setup
• Use Gemini 4 Hydrogen in Cherry Studio: OpenAI Chat Completions · base URL and API key setup
• Use Gemini 4 Hydrogen in WorkBuddy: OpenAI Chat Completions · base URL and API key setup
• Use Gemini 4 Hydrogen in OpenClaw: OpenAI Chat Completions · base URL and API key setup
• Use Gemini 4 Hydrogen in Hermes Agent: OpenAI Chat Completions · base URL and API key setup
• Use Gemini 4 Hydrogen in Magpie: OpenAI Chat Completions · base URL and API key setup
• OpenRouter alternative: Wokey vs OpenRouter: Compare token prices, payment fees, API forms, and upstream sources.
• Verifiable AI API: which models and API forms carry response proofs: Response proofs currently cover only raw passthrough Claude Messages and GPT Responses on official routes.

Frequently asked questions

Related models and documentation

• Gemini 4 Argon
• Gemini 4 Carbon
• Gemini 3.8 Flash
• API documentation
• All models and pricing
• Provider Node