Gemini 4 Hydrogen pricing: $1.50/$7.50, 1M context and the 3.8 Flash switch

Summary. Google released Gemini 4 Hydrogen (gemini-4-hydrogen) on 12 November 2026 as a generally available model on the free and paid tiers of the Gemini API and Google AI Studio. The paid rate is $1.50 per million input tokens and $7.50 per million output tokens, flat across a shared 1,048,576-token window, with implicit cache hits at $0.075. Google names it the recommended upgrade from Gemini 3.8 Flash, which costs an introductory $0.75 and $3.75 until 31 December and moves to the same $1.50 and $7.50 on 1 January 2027. Hydrogen drops explicit caching, defaults to the high thinking level, adds xhigh and caps each synchronous response at 65,536 tokens unless you use background long decode. GitHub made it generally available in Copilot on 13 November and removes 3.8 Flash from Copilot on 14 December. In our worked example, a support assistant that costs $6,290 a month on 3.8 Flash in January costs $5,775 on Hydrogen if cache hits hold, or $7,440 if they fall and thinking runs long. Test now, keep 3.8 Flash at half price through December, cut over before 1 January, and set store explicitly, because new Gemini API projects log prompts for 55 days by default.

About this article. This news analysis is for engineering leads, FinOps owners and CTOs in the US and UK who run Gemini 3.8 Flash or plan Gemini API budgets. We checked every price, model ID, limit and date on 15 November 2026 against Google's Gemini API release notes, pricing, model, migration, deprecation and logging pages, Google's launch post, Google Cloud's Agent Platform pricing page, GitHub's changelog and Artificial Analysis. Vendors change prices often, so re-check before you commit; nothing here is legal advice, and no vendor paid for inclusion. Researched with AI assistance and reviewed by our editorial team.

What Google shipped on 12 November, and what GitHub did a day later

Google's release notes list Hydrogen as generally available on 12 November 2026. The launch post, by Koray Kavukcuoglu of Google DeepMind, positions it as the fast model for everyday coding and agent work in the Gemini 4 family, next to the frontier models Gemini 4 Carbon and Gemini 4 Argon. It ships under a stable model ID with no preview suffix. Google's What's new in Gemini 4 Hydrogen guide lists every API behaviour that differs from 3.8 Flash.

Three points matter for buyers. First, the price is flat across the whole million-token window, so there is no long-context surcharge to model. Second, Hydrogen is now the default model for the Antigravity agent in Gemini Managed Agents, which shows where Google is steering agent workloads. Third, Google says Hydrogen beats Argon on the Artificial Analysis Intelligence Index at its default setting (51 against 50, and 54 at xhigh), more than triples 3.8 Flash's Terminal-Bench 4.1 score (57% against 18.2%) and cuts fabricated claims by more than five times. Artificial Analysis's own model page matches the index figures and adds the lower settings: 41 at medium and 27 at low, against 31 for the best 3.8 Flash setting. The Terminal-Bench and fabrication figures are Google's own. Treat them as a reason to test, not as a result.

On 13 November, GitHub made Hydrogen generally available in Copilot for the Pro, Pro+, Max, Business and Enterprise plans, billed at the provider's list price under usage-based billing. A second changelog entry the same day deprecates Gemini 3.8 Flash in every Copilot experience on 14 December 2026, with Hydrogen as the suggested alternative. Gemini 4 Argon leaves Copilot that day too, replaced by Carbon.

Gemini 4 Hydrogen vs Gemini 3.8 Flash: price, limits and behaviour

On the Gemini API itself, 3.8 Flash is not being shut down. The deprecations page lists gemini-3.8-flash, released on 2 September 2026, with no shutdown date announced. The forcing event is the price step on the pricing page, which we cover in our Gemini 3.8 Flash price increase guide. The comparison below comes from Google's pricing page and the two models' documentation.

Neither model accepts the minimal thinking level; both return an error. From January the headline rates match, and the bill difference comes from three places.

Caching. Hydrogen's cached input costs half of 3.8 Flash's January rate ($0.075 against $0.15), with no hourly storage charge. Implicit caching applies only to prompts of at least 4,096 tokens with a matching prefix, and Google says it carries no guaranteed saving, unlike explicit caching. In multi-turn agent loops on Gemini 4, the previous request is cached only when the new turn adds at least 2,050 tokens. A team that pinned a large explicit cache on 3.8 Flash gives up a guaranteed discount for a probable one, and prompt order now matters more than it did.

Thinking tokens. Hydrogen defaults to high, where 3.8 Flash defaults to medium, and every thinking token bills at the $7.50 output rate. Google says partners saw 12% to 23% lower token use than 3.8 Flash in early testing, but on their own workloads. Set thinking_level yourself and measure.

Output ceilings. Because Hydrogen shares one window across input, thinking and output, Google tells you to remove hardcoded low max_output_tokens caps such as 8,192 and to use background execution for anything longer than 65,536 tokens.

Below Hydrogen, Google still recommends Gemini 3.5 Flash-Lite ($0.30 input, $2.50 output) for high-volume classification and routing. For the hardest long-horizon reasoning, Google points to Carbon and Argon; our Gemini 4 Carbon pricing article covers Carbon's move to $4 and $20 on 1 December.

Why switching production in November costs more

Until 31 December, Gemini 3.8 Flash costs exactly half of Hydrogen on every paid tier. A team that moves all production traffic to Hydrogen now pays double rates for the rest of the year, about six and a half weeks. Waiting until 1 January to start testing leaves no buffer if the migration finds a problem.

A sensible sequence for most Gemini API customers:

1. Now to early December: run Hydrogen on shadow or canary traffic (5% to 10%), fix the API changes listed below, and measure tokens per task at medium and high.

1. Mid-December: decide per workload. Where Hydrogen's quality gain is clear, switch then and accept a few weeks at the higher rate. Where it is a wash, keep 3.8 Flash until the price step.

1. By 31 December: complete the cut-over, or knowingly accept 3.8 Flash at $1.50 and $7.50 from 1 January with explicit caching still available.

GitHub Copilot users do not get that choice. 3.8 Flash leaves Copilot on 14 December, and under default model enablement Hydrogen is switched on automatically for Business and Enterprise customers unless an administrator turned the global default off or disabled the model; in that case, enable it through its model policy before 14 December.

Two automatic moves matter too. Google's deprecations page says gemini-3.7-flash already routes to gemini-3.8-flash, and the 2 November release note says gemini-2.5-pro requests route to 3.8 Flash from 1 December 2026, billed at 3.8 Flash rates; API access to 2.5 Pro is already limited to existing users. Code on either string lands on a model that doubles in price a month later, so it is often simpler to move straight to Hydrogen.

Worked cost example: a 150-person US SaaS support assistant

Assume a 150-person US B2B software company runs an in-app support assistant on the paid Gemini API. Monthly volume is 4,000 million input tokens and 350 million output tokens including thinking. Half of the input is a repeated prefix: system prompt, policy text and product docs. On 3.8 Flash the team holds a 500,000-token explicit cache for the whole month (730 hours, so 365 million token-hours of storage). Swap in your own volumes; our LLM token counter guide explains how to count them. All rates come from Google's pricing page.

When the cache hit rate holds and thinking stays at the old level, Hydrogen costs $515 a month (8.2%) less than 3.8 Flash from January. If implicit caching catches only 30% of input and the high default adds 20% to output, Hydrogen costs $1,150 a month (18.3%) more.

Now apply two tuning steps: set thinking_level to medium for chat turns, assuming output then stays at 350 million tokens, and move the nightly article-enrichment job (a quarter of all tokens) to the Batch API. For a fair test, 3.8 Flash gets the same Batch split at January rates, keeping its explicit cache.

Tuned Hydrogen is $1,236.87 a month below 3.8 Flash's untuned January bill, about $14,800 a year, and $496.25 (about 9%) below 3.8 Flash with the same Batch split. It is still $1,908.13 above the December bill. Our LLM hybrid routing framework covers the next step, sending simple turns to a cheaper model such as Flash-Lite.

Decision table: stay, switch or split

Each row rests on facts cited above; Google's pricing page lists the Gemini 3.8 Live models.

Privacy check: new Gemini API projects store prompts by default

Since 26 October 2026, paid-tier projects created on or after that date store Generate Content API requests (generateContent, streamGenerateContent and batchGenerateContent) by default, with store set to true. Older projects keep storage off unless someone enabled logging in AI Studio. The Interactions API, which Google's Hydrogen quickstart uses, stores requests by default on every project. A fresh project for your Hydrogen rollout logs full prompts and responses unless you say otherwise.

Google's logging pages say:

• Retention. Logs are kept for 55 days by default. A project owner can set 7, 14, 28 or 55 days. Logs saved into a dataset do not expire.

• Training use. Logs from billing-enabled projects are not used for product improvement unless you share a dataset with Google. Shared datasets fall under Google's data-use terms for unpaid services, and human reviewers may read them.

• UK and EEA users. Google tells developers not to donate data from API clients available to users in the European Economic Area, Switzerland or the United Kingdom.

• Contributor Tier. An invite-only private preview offers Hydrogen at $0.75 and $3.75 (Google says up to 50% off) in exchange for using your data under the unpaid-services terms. Do not enrol a project that handles customer or employee data.

• Free tier. Google states that free-tier content is used to improve its products, which rules it out for personal or confidential data.

How to turn it off for a request:

            from google import genai

client = genai.Client()  # google-genai 2.33.0 or later for the store field
prompt = "Summarise this support ticket and suggest a reply."

response = client.models.generate_content(
    model="gemini-4-hydrogen",
    contents=prompt,
    config={
        "store": False,  # overrides the project-level logging default
        "thinking_config": {"thinking_level": "medium"},
        "max_output_tokens": 65536,
    },
)
          

The request-level store field overrides the project setting. It needs google-genai 2.33.0 or later for Python, @google/genai 2.31.0 or later for JavaScript, google-genai 1.79.0 or later for Java and google.golang.org/genai 1.76.0 or later for Go. You can also switch Generate Content logging off for the whole project in the Settings panel of the AI Studio Logs dashboard.

There is one trade-off. Google warns that turning Interactions API logging off stops the API from storing and retrieving conversation history, and Hydrogen's migration guide recommends previous_interaction_id for multi-turn tool use. If your data rules require no storage, manage history on the client (stateless mode) and resend every thought block exactly as received. For UK teams, a move from no storage to 55 days of full prompt logging is worth recording in your UK GDPR records of processing and any data protection impact assessment.

Migration steps from Gemini 3.8 Flash to Hydrogen

Several items in Google's checklist fail quietly rather than with an error.

1. Find every model string. Search code, config and gateway rules for gemini-3.8-flash, gemini-3.7-flash and gemini-2.5-pro.

1. Set logging deliberately. Check each project's creation date and settings panel, then set store per request where it matters.

1. Change the model ID to gemini-4-hydrogen. Carbon and Argon users need no request schema change.

1. Remove explicit caching. Delete cachedContents and caches.create calls and any cached_content field. Put static content first in each prompt so implicit caching can match the prefix; it applies only to prompts of at least 4,096 tokens.

1. Set the thinking level. Hydrogen defaults to high. Run your evaluation set at low, medium, high and xhigh, then pick the lowest level that passes.

1. Fix output limits. Remove low max_output_tokens caps or raise them to 65,536. For longer outputs, set background to true and max_output_tokens above 65,536, or use client-side continuation.

1. Drop sampling parameters. Hydrogen ignores in-range temperature, top_p and top_k and returns a 400 error for out-of-range values.

1. Declare tools natively. Prompting the model to emit raw-text JSON tool calls with no declared tools can return MALFORMED_FUNCTION_CALL. Use function calling or structured outputs.

1. Preserve state. Use previous_interaction_id, or in client-managed histories return prior turns and thought signatures unchanged. Do not copy one signature across parallel function calls.

1. Re-baseline tokens and cost. Google bills every token processed, including thinking tokens and the internal tool instructions and function declarations added when grounding is on. Compare cost per completed task, not per request, over at least a week of shadow traffic.

1. Cut over with a rollback. Keep the 3.8 Flash path behind a flag until January. It stays on the Gemini API with no announced shutdown.

Teams that run Gemini behind a gateway can do steps 1, 3 and 11 centrally; our AI gateway comparison covers the routing options.

Platform notes for US and UK teams

Agent Platform. Google's launch post says enterprises can deploy Hydrogen through Gemini for business. Google Cloud's Agent Platform pricing page now lists Hydrogen at $1.50 input and $7.50 output on the global endpoint, $1.65 and $8.25 on regional endpoints, and $0.075 for cache reads, with Batch at $0.75 and $3.75 and Priority at $2.70 and $13.50. Two notes on that page matter. A cache-write charge, $1.875 per million tokens on the global endpoint, applies from 12 January 2027. And since 13 November, Gemini 4 completions longer than 32K tokens are billed for retrievable token usage even if the request later fails on a timeout. If your workloads run through Google Cloud billing rather than an AI Studio key, confirm Hydrogen's availability in your regions before you plan the move.

What is still unconfirmed. Google has not published a shutdown date for 3.8 Flash on the Gemini API, an independent benchmark of Hydrogen against 3.8 Flash on business workloads, or any guaranteed implicit-cache hit rate. Plan around all three.

How eCorpIT can help

eCorpIT is a Gurugram-based technology organisation, founded in 2021, assessed at CMMI Level 5 and MSME certified, with senior-led engineering teams working across AWS, Microsoft and Google platforms. Our AI and machine learning services team can inventory your Gemini model strings, build an evaluation set from real traffic and compare Hydrogen with 3.8 Flash on cost per completed task before anything reaches production. Talk to us at /contact-us/.

Last updated: 15 November 2026.

References

1. Gemini API release notes (12 November, 2 November and 26 October 2026 entries) — Google AI Studio

1. Gemini Developer API pricing — Google AI Studio

1. What's new in Gemini 4 Hydrogen — Google AI Studio

1. Gemini 4 Hydrogen in GitHub Copilot, 13 November 2026 — GitHub Changelog

1. Upcoming deprecation of selected GitHub Copilot models in December, 13 November 2026 — GitHub Changelog

1. Logs and datasets — Google AI Studio

1. Introducing Gemini 4 Hydrogen, 12 November 2026 — Google

1. Gemini 4 Hydrogen model page — Google AI Studio

1. Gemini 4 Hydrogen: intelligence, performance and price comparison — Artificial Analysis

1. Gemini deprecations — Google AI Studio

1. Gemini 3.8 Flash model page — Google AI Studio

1. Context caching — Google AI Studio

1. What's new in Gemini 3.8 Flash — Google AI Studio

1. Long decode — Google AI Studio

1. Data logging and sharing — Google AI Studio

1. Gemini thinking — Google AI Studio

1. Agent Platform pricing — Google Cloud