Gemini 4 Hydrogen launches with free API access and $1.50 input pricing
Google released Gemini 4 Hydrogen on November 12, adding a faster-execution option to its Gemini 4 family. The practical news is access: developers get a free API tier, while standard paid usage starts at $1.50 per million input tokens. Its large context window comes with an important qualification: input, reasoning and output share the same budget.
Where you can use Hydrogen
Google’s launch announcement positions Hydrogen for everyday coding and agent workflows, with Carbon and Argon handling deeper reasoning. Rollout includes AI Studio, the Gemini API, Gemini for business, the Gemini app and Search. Antigravity access is for paid Google AI Pro and Ultra subscribers. In the Gemini app, free users receive Hydrogen through automatic model selection when appropriate, rather than a guaranteed manual model choice.
The developer guide confirms the model ID gemini-4-hydrogen, free and paid tiers, and both GenerateContent and Interactions API support. You can open the model in Google AI Studio. Free API access remains rate-limited.
Update, November 13, 2026: Vercel’s November 12 announcement adds another access route: AI Gateway, using the model ID google/gemini-4-hydrogen. Vercel provides an AI SDK example and coding-agent setup instructions, plus usage and cost tracking and configurable retries and failover. Keep this gateway route separate from direct Google API access when comparing billing and limits; do not assume Google’s free tier or Cloud timeout policy applies unchanged.
Pricing: thinking tokens count
The standard pricing table lists $1.50 per million input tokens, $7.50 per million output tokens including thinking, and $0.075 per million implicitly cached input tokens. Explicit caching is unsupported. Google lists free-tier data as used to improve its products and paid-tier data as not used for that purpose. Search and Maps grounding are unavailable on the free tier and carry separate paid-tier allowances and charges.
For a concrete budget, 100,000 uncached input tokens plus 10,000 billed output tokens cost $0.225 at standard rates, before tool charges. That is arithmetic, not a measured task cost. An agent that retries, reasons longer or makes several calls can spend more even when its final answer is short.
Google Cloud timeout billing changes on November 13
Update, November 12, 2026: Google Cloud’s pricing page says a Gemini 4 exception takes effect on November 13, 2026: when a completion exceeds 32K tokens, retrievable token usage is billable even if the request ultimately fails because of a client or server timeout. The page gives no time or timezone.
This qualifies the page’s general rule that non-200 responses incur no input or output charges. It does not say every failed request is billable. This is Google Cloud guidance, not confirmation of the same policy for the Gemini Developer API. For long Cloud jobs, check retrievable usage before retrying rather than assuming a timeout means no charge.
The 1M-token window is shared, not doubled
The model specifications list a unified 1,048,576-token context window and a 65,536-token default synchronous output limit. Inputs can include text, images, video, audio and PDFs. Output is text. Image generation, audio generation and the Live API are not supported.
The developer guide clarifies that input, thinking and output all consume that unified window. Going beyond a 65,536-token response requires background execution or client-side continuation. Setting background execution alone does not raise the limit: the output cap also needs to increase. This is not a promise of a million input tokens plus another million output tokens in one request.
Migration requires more than a new model name
Hydrogen offers low, medium, high and xhigh thinking levels, with high as the default. Google’s migration guide says minimal is rejected, custom sampling parameters are ignored, and explicit cache fields must be removed. Existing applications should preserve conversation state and thought signatures and declare tools or structured-output schemas natively.
For a first evaluation, keep the same task, tools and acceptance tests across thinking levels. Record complete-task cost, latency and reviewer corrections. A lower setting is useful only if it still meets the quality bar. ByteForward has not run hands-on performance tests of Hydrogen.
A faster partner for Carbon, not a universal replacement
Google also announced an Antigravity Orchestrator that switches models within a session. It reports near-Carbon task success on Terminal-Bench 4.0 with 24% less quota and 19% faster completion. Those are Google’s results for that configuration, not a guaranteed saving on your project.
For teams already using Gemini 4 Carbon, the useful comparison is which stages justify deeper reasoning and which can finish reliably at lower cost. Start with a bounded coding or document task rather than moving every workflow at once.
AI-generated editorial illustration for ByteForward, not a Google product or interface. Logo reference: official Gemini website. Gemini is a trademark of Google LLC. ByteForward is not affiliated with or endorsed by Google.
