Google Launches Gemini 4 Hydrogen at $1.50 per Million Input Tokens

Google launched Gemini 4 Hydrogen on Nov. 12, a model built for fast coding and agent-style work that is rolling out across its developer, business and consumer products, including the Gemini app and Google Search. Free Gemini app users will get it through automatic model selection when Google considers it appropriate for a prompt, according to Unite.AI.

Google says the model is faster and cheaper to run than its top-tier Gemini 4 models, though one comparison in its own model card shows a rival ahead on a coding test.

The key numbers

$1.50 and $7.50: API pricing per million input tokens and per million output tokens, applied flat across the entire 1 million-token context window, Google said.
2.5 times faster: how much faster Hydrogen finishes tasks than Gemini 3.8 Flash at its default high thinking setting, according to Google.
57% vs. 18.2%: Hydrogen’s Terminal-Bench 4.1 score compared with Gemini 3.8 Flash.
51 vs. 50: Hydrogen’s score on the Artificial Analysis Intelligence Index at high thinking versus Gemini 4 Argon’s, rising to 54 at the xhigh setting.
24% and 19%: the average reduction in quota use and the average speedup that Google’s Antigravity team reported for the new Orchestrator tool on a Terminal-Bench evaluation.
63.9% vs. 57.0%: Unite.AI noted that the model card’s cheating-adjusted Terminal-Bench 4.1 table lists Claude Sonnet 5.5 above Hydrogen.

What is behind the numbers

Hydrogen sits below Gemini 4 Carbon and Gemini 4 Argon, which Google describes as better at deep, long-running reasoning. Hydrogen is meant for quick execution such as writing code, running tests and verifying changes, either on its own or paired with the larger models.

Orchestrator, a new option in Google Antigravity, applies that split inside one session. As Unite.AI described it, Carbon can plan in a read-only phase, Hydrogen then edits code and runs tests, and control can return to Carbon if execution hits trouble. Google said Orchestrator reached near-Carbon task success.

The model has four thinking levels: low, medium, high, which is the default, and a new xhigh tier. Google said it also cut fabricated claims by more than 5x compared with Gemini 3.8 Flash. Unite.AI reported that the model card records a 5.57-percentage-point increase in unjustified refusals versus Gemini 3.8 Flash, and lists hallucinations, occasional slowness or timeouts, and potentially higher token use at elevated thinking levels among the model’s limitations.

What this means for users and developers

For everyday users, Google says the Gemini app follows tone and formatting instructions more closely and picks tools more reliably, while Search handles multipart or typo-filled questions better and gives more visual answers. Paid Google AI Pro and Ultra subscribers can also use Hydrogen in Antigravity.

For developers, the Gemini API documentation lists a rate-limited free tier alongside the paid pricing, and cached input at $0.075 per million tokens, Unite.AI reported. Orchestrator is initially rolling out to AI Ultra subscribers, with enterprise access to both through Gemini for business.

Google said that in a comparison with Gemini 3.8 Flash, its model card reports reductions of up to 81% in agentic overreach and 79% in under-service. Whether the speed and cost gains hold up in everyday use, alongside the documented limitations, remains to be seen.

Around Prague, based on reporting by blog.google, Unite.AI, The American Bazaar

Cover photo: Asoundd / Wikimedia Commons, CC BY-SA 4.0, cropped

Share: Telegram WhatsApp