Updated on November 14, 2026
Gemini 4 Hydrogen is Google DeepMind's fast execution model for everyday coding, document analysis and agentic workflows. Announced on November 12, 2026 as the third member of the Gemini 4 family after Argon and Carbon, it replaces Gemini 3.8 Flash. The API charges $1.50 per million input tokens and $7.50 per million output tokens with a rate-limited free tier on Google AI Studio; Google Search uses it for all queries worldwide at no cost. Picking Hydrogen by name in the Gemini app starts at $4.99/month on Google AI Plus.
Hydrogen finishes tasks roughly 2.5 times faster than Gemini 3.8 Flash at its default high thinking level, according to Google. Artificial Analysis gives it a 51 on its Intelligence Index at high and 54 at xhigh, compared with 50 for Argon and 55 for Claude Sonnet 5.5 at maximum. The cost per finished index task comes in at about $1.87 at high and $3.16 at xhigh, roughly a third of what Sonnet 5.5 costs for a comparable score.
Inputs can be text, images, audio, video or PDF, all within a one-million-token window; output is text only, up to one million tokens. The knowledge cutoff is March 2026. On the safety side, Google reports an 81% reduction in agentic overreach and 2.3 times better resistance to indirect prompt injection on the Gray Swan benchmark.
| Benchmark (source: Google) | Gemini 4 Hydrogen | Claude Sonnet 5.5 | GPT-6.1 Terra |
|---|---|---|---|
| Agents' Last Exam v1.1 | 71.6% | 60.4% | 69.8% |
| AutomationBench-AA | 76.1% | 71.8% | 60.2% |
| Terminal-Bench 4.1 | 57.0% | 63.9% | not reported |
| HLE-Diamond (reasoning) | 58.8% | 53.3% | 47.3% |
| MKA (multimodal) | 53.4% | 52.0% | 43.8% |
Inside Google Antigravity, Ultra subscribers can turn on Orchestrator, which pairs Carbon (for planning) with Hydrogen (for execution) in a single session. On Terminal-Bench 4.0, Google says this duo reached near-Carbon success rates while consuming 24% less quota and finishing 19% faster.
Teams still on Gemini 3.8 Flash see a dramatic jump: Terminal-Bench 4.1 climbs from 18.2% to 57%, and fabricated claims drop by more than five times. The per-token price doubles (from $0.75 to $1.50), though Google will raise Flash to the same rate on January 1, 2027, making the upgrade cost-neutral. Worth noting: some API parameters differ, so test before you swap a production pipeline.
The Standard API tier bills $1.50 input and $7.50 output per million tokens, thinking tokens included. Batch and Flex cut those figures in half ($0.75 / $3.75) for deferred jobs. Implicit caching drops the input rate to $0.075 per million tokens, a 95% discount.
For consumers, Hydrogen already powers Google Search for everyone. In the Gemini app, free accounts may receive it through Auto model selection, but most prompts still route to Flash-Lite. The model appears in the picker starting at Google AI Plus ($4.99/month); Pro ($19.99) adds Sheets and Antigravity access, and Ultra unlocks the highest limits and Orchestrator. As with any large language model, pricing may change; Google's official page is the definitive reference.
Partly. The Gemini API and Google AI Studio have a rate-limited free tier, and Google Search runs Hydrogen for everyone at no charge. In the Gemini app, free accounts only reach it through Auto model selection; most prompts go to Flash-Lite. Picking it manually starts at $4.99/month on Google AI Plus.
It depends on the workload. Google's own benchmarks show Hydrogen ahead on automation, multimodal knowledge work and reasoning tasks, at roughly a third of Sonnet 5.5's cost per task according to Artificial Analysis. Sonnet 5.5 still leads on Terminal-Bench 4.1, OSWorld-2.1 and terminal-based science. Choose based on your primary task profile.
Hydrogen accepts text, images, audio, video and PDF within a one-million-token input window. Output is text only, also up to one million tokens. Image generation, audio generation and the Live API are not supported according to Google's model page.
Yes. Google's pricing page states that content sent on the free tier may be used to improve its products. Paid tiers (Standard, Batch, Flex, Priority) do not use your data this way. If your prompts contain sensitive or customer-facing information, the paid tier is the safer choice.
Verdict: At about $1.87 per finished task and an AA score of 51 at its default setting, Gemini 4 Hydrogen fits developers and product teams looking for a fast, affordable model for routine coding, document synthesis and multi-step agents. For heavy terminal work or science-focused reasoning, pairing it with Carbon or a rival frontier model fills the gap.
« An all-in-one creative platform that brings together more than 30 image and video models in a single interface. It includes camera controls that let you direct like a real filmmaker, characters that remain consistent from scene to scene, and ready-to-use tools for ads and short-form content »
Claude Fable 5.2 
Claude Opus 5.5 
Gemini 4 Carbon 
