Confirmed 12 Nov 2026 · Labs A one-minute read

Google released Gemini 4 Hydrogen, its new model for coding and agent workflows.

The successor to Gemini 3.8 Flash shipped on November 12 across Google's developer and business products. Reported pricing is $1.50 per million input tokens and $7.50 per million output tokens, with a rollout into Search.

Until now

The fast tier was the cheap tier, for high-volume everyday work.

Now

The fast tier is priced as a product, not a loss leader.

Why it matters

Flash is how Google held the floor of the market. Hydrogen is faster, scores higher, and is reported to be the model Search will answer with for every user. If the volume tier now carries mid-frontier prices, every product costed on cheap tokens has to be recosted.

Established

Google released Gemini 4 Hydrogen on November 12, succeeding Gemini 3.8 Flash.
It is available in the Gemini API, Android Studio, Antigravity and Gemini for business.
Google says it has tighter instruction following and fewer fabricated claims.

Not established

The pricing. Google's own post gives no rates; $1.50 and $7.50 are reported.
The benchmark scores. Every figure published is Google's own, untested here.
That Search answers change. The rollout is announced, not measured.

The five-minute readThe record, the timeline, the weight and 4 sources

The record

Google published "Gemini 4 Hydrogen: everyday coding and agentic workflows for developers" on blog.google on November 12, 2026, with page metadata of 18:00 Universal Time. We read it. Google describes Hydrogen as a model for "everyday coding and agentic workflows," usable at 4 configurable thinking levels or alongside Gemini 4 Carbon and Gemini 4 Argon, and claims "tighter instruction following and fewer fabricated claims." The post names availability in the Gemini API in Google AI Studio, Android Studio, Gemini for business, and Google Antigravity for paid AI Pro or Ultra subscribers. It gives no pricing, no benchmark figures and no context window. VentureBeat, published at 2:30 pm Pacific Time, reports standard rates of $1.50 per million uncached input tokens, $0.075 per million cached input tokens and $7.50 per million output tokens including reasoning tokens, with Batch, Flex and Contributor tiers at half those rates, and says the uncached input and output rates are 15 times the lowest standard rates from Claude Haiku 5.5, GPT-6 Luna and Meta's Muse Ember. It reports Hydrogen completing tasks 2.5 times as fast as Gemini 3.8 Flash at default reasoning, scoring 51 on the Artificial Analysis Intelligence Index at high reasoning against 31 for Flash, and 57% on Terminal-Bench against 18.2% for Flash, with Claude Sonnet 5.5 at 63.9% in Google's own table. Box is reported at 79% accuracy against 74% with Flash, in less than half the time and about 12% fewer tokens. A second summary of Google's material gives a 1,048,576-token context window, thinking levels of low, medium, high and xhigh, 58.8% on HLE-Diamond, and a Search rollout to all users in all AI Mode languages. Nothing beyond the launch and the surfaces is in Google's own post.

How it got here

Sep 30
Google releases Gemini 4 Argon to cyber defenders first and paying customers later. Edition 038 carded it.
Nov 9
Google launches 2 AI chips, one built to protect model weights while in use. Edition 083 carded it.
Nov 11
Google ships a model that only decides, priced at $0.10 per million input tokens. Edition 088 carded it. The cheap end of the range was 1 day old.
Nov 12
Gemini 4 Hydrogen ships as the successor to Gemini 3.8 Flash, with standard rates reported at 15 times the cheapest rivals' and a Search rollout to all users reported.

Who has to react

Anthropic, OpenAI and Meta, whose cheapest tiers are the comparison; every developer who budgeted on Flash rates; Box and the other early-access testers; Google Search advertisers, whose answers Hydrogen will write; the benchmark publishers Google cites; Google Cloud's enterprise customers.

What to watch

Would strengthen itAnthropic, OpenAI or Meta raises the price of its cheapest standard model tier, or retires that tier without a cheaper replacement, by March 31, 2027.
Would weaken itGoogle cuts Gemini 4 Hydrogen's standard input or output rates, or ships a cheaper general-purpose tier below $0.50 per million input tokens, by June 30, 2027.
Would strengthen itGoogle states in a post, filing or earnings call that Gemini 4 Hydrogen serves AI Mode or AI Overviews in Search, by December 31, 2027.

Why it weighs this much

Signal strength

CostLow

CommitmentHigh

ReversibilityHigh

PersistenceQuarter

Cost is low. Shipping a model into surfaces Google already owns costs nothing it has not already spent, and the serving bill is paid in compute it built. Commitment is high in a narrow sense: a published model with a published name and rates becomes a price other people plan against, and Google cannot quietly unship Search. Reversibility is high, because rates and model defaults change with a configuration file, and Google has cut prices before. Persistence is a quarter: this generation turns over fast, and Edition 088's decisions model was 1 day old when this landed.

The words

Flash tier: Google's name for its fast, cheap model line, used where volume matters more than depth.
Thinking level: A setting controlling how much reasoning a model does before answering. Higher levels cost more, because reasoning tokens are billed as output.
Cached input: Repeated prompt text the provider stores and charges less for. Reported at $0.075 per million tokens, 20 times cheaper than fresh input.
Context window: How much text a model holds at once. Hydrogen's is reported at 1,048,576 tokens, shared across input, thinking and output.

Sources

Your call

Which explanation do you believe?

A call is final. Other people's calls show after yours.