Gemini 4 Hydrogen Overview: Release, API and Pricing

Hydrogen, Carbon and Argon: what happened to Flash?

Google positions Hydrogen for rapid execution, Carbon for the hardest reasoning and verification, and Argon for advanced long-horizon work with lower token use than Carbon. Hydrogen can work alone or handle worker and verifier turns while Carbon handles planning and difficult roadblocks.

Hydrogen is the English name of the element hydrogen. The name alone does not specify model size, architecture or capabilities. Google's migration guide explicitly recommends it for users of Gemini 3.8 Flash and earlier models; that is stronger evidence than guessing a Pro-to-Argon or Flash-to-Hydrogen naming rule. Flash-Lite remains a separate low-cost option. Do not assume every Gemini product has been renamed.

At launch Google reported an Artificial Analysis Intelligence Index score of 51 at default high thinking, versus Argon's 50, rising to 54 at xhigh. This does not make Hydrogen universally stronger than Carbon or Argon. Thinking level, tools, latency and task mix matter. See the Carbon overview and Carbon vs Argon.

Gemini 4 Hydrogen context, output and thinking levels

The 1,048,576-token window is shared by input, thinking and output. It is not a separate 1M input allowance plus another 1M output allowance. Default synchronous output is 65,536 tokens per hop. Longer generation requires background execution or continuation and must fit the remaining shared context.

Hydrogen supports low, medium, high by default, and xhigh. Start at high for difficult coding, then check whether medium or low meets your acceptance criteria for repeated work. Reserve xhigh for cases where deeper reasoning improves the accepted result enough to justify extra output tokens and waiting.

Google documents function calling, structured outputs, code execution, computer use, Search/Maps grounding and implicit caching. It does not list image generation, audio generation or the Live API as supported. Reading video does not mean generating video. Provider-compatible routes can expose fewer tools than Google's native API.

What do the Hydrogen benchmarks actually show?

Google's published evaluation table gives a mixed picture. These are Google's reported results and cited leaderboard numbers, not APIMaster measurements.

Hydrogen leads these SWE-sweep, knowledge-work and reasoning rows, while Sonnet leads Terminal-Bench 4.1 and the selected computer-use test. An unreported Terra score is missing data, not zero.

The evaluation methodology specifies different harnesses and aggregation: SWE-sweep uses Antigravity CLI 1.4.8; Terminal-Bench 4.1 uses high effort and averages five attempts per task; OSWorld uses an 82-task offline subset and a maximum over three runs. Most other results use the highest thinking setting unless specified. Scores from these rows cannot be averaged into a universal win rate.

Gemini 4 Hydrogen API pricing and the December comparison

Hydrogen's current Standard $1.50 / $7.50 applies across its full context window. Google also lists $0.75 / $3.75 for Batch and Flex, $0.075 cached input through implicit caching, and a rate-limited free tier. Those are separate billing conditions; free access is not an unlimited production SLA.

Carbon and Argon currently cost $2 / $10, scheduled to become $4 / $20 on December 1, 2026. Hydrogen is therefore 25% cheaper per token than their current Standard rates and 62.5% cheaper than those scheduled December rates. It would cost 37.5% of their price, not “37.5% cheaper.” Google's 3.8 Flash promotion is $0.75 / $3.75 through December 31, scheduled to match Hydrogen's current rate on January 1, 2027.

These are Google's published schedules, not a guarantee that Hydrogen can never change price. A newer model is also not automatically cheaper: Hydrogen currently costs twice Google's promotional 3.8 Flash rates for the same token counts. Thinking, tools, cache hits and completed-task rates can change the actual bill.

How to use the Hydrogen API and migrate from 3.8 Flash

Google's native API is available today. A minimal Interactions request is:

curl "https://generativelanguage.googleapis.com/v1beta/interactions" \
  -H "x-goog-api-key: $GEMINI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gemini-4-hydrogen","input":"Review this retry-queue design and identify race conditions.","generation_config":{"thinking_level":"high","max_output_tokens":65536}}'

This is an example, not a request we executed. Its token limit caps combined thinking and output; it is a ceiling, not a fixed bill. Google's key and base URL are separate from APIMaster credentials.

For a 3.8 Flash migration, update the model ID, remove explicit cached_content usage, and preserve thought signatures and prior turns. Hydrogen uses implicit caching only, with a documented 4,096-token minimum for eligible prompts. minimal thinking is rejected; custom temperature, top_p and top_k values are ignored. Declare native tools or structured output schemas rather than asking for improvised text-form tool calls.

Keep saved prompts and acceptance tests, then compare completion rate, human corrections, latency and all billed tokens. Google's claim of 2.5 times faster task completion than 3.8 Flash concerns its reported evaluation, not every workload or APIMaster route.

APIMaster access and related guides

Hydrogen is not yet available on APIMaster. Check current prices and availability on the model card. Discounted routes are rolling out.

For available Gemini access: register → create an API key → use https://apimaster.ai/v1 with an enabled model such as gemini-4-carbon or gemini-3.8-flash after checking its live card. Hydrogen's official model ID is gemini-4-hydrogen; its APIMaster route price remains unannounced at this check.

Continue with the Hydrogen buying guide, three-model comparison, or existing Gemini Flash price guide.

FAQ

When was Gemini 4 Hydrogen released?

Google's Gemini API changelog dates general availability to November 12, 2026. It is a released API model, not a rumored codename awaiting a date.

Is Hydrogen the same as Gemini 4 Flash?

The official ID is gemini-4-hydrogen. Google recommends it as the upgrade from 3.8 Flash and describes it as the fast everyday Gemini 4 model. Use the official ID rather than inventing gemini-4-flash as an alias.

Is Gemini 4 Hydrogen free?

Google lists a rate-limited free API tier. Paid Standard costs $1.50 input and $7.50 output per million tokens. Free-tier terms and product subscriptions are separate from paid API routes.

Does Hydrogen support one million output tokens?

Only through long decode within the remaining shared context. Ordinary synchronous generation defaults to a 65,536-token hop; input, thinking and output share the total 1,048,576-token window.

Can I buy discounted Hydrogen on APIMaster now?

No public APIMaster route was listed at the November 13 check. Discounted access is being rolled out; existing Carbon, Argon and Flash quotes must not be presented as Hydrogen prices.

Sources

• Google launch announcement and API release notes.
• Hydrogen model specification, migration guide and current pricing.
• Google evaluation methodology and results.
• APIMaster marketplace, OpenRouter catalog and Vercel catalog, checked November 13, 2026.