Oleksandr Kuzmenko, EditorPublished 6 min readAI-assisted · editor-reviewed
TL;DR
- Gemini 4 Hydrogen is Google's new entry-level model powering the free tier and Google Search AI features.
- Hydrogen is built for fast implementation, diff verification, and test execution rather than deep architectural reasoning.
- Hydrogen currently launches at only a 25% discount to Carbon and Argon, widening to 75% when their prices double on December 1st.
- Hydrogen costs roughly 15x more than entry competitors Claude Haiku 5.5 and GPT-6 Luna.
Key facts
- Hydrogen Launch Discount
- 25% below Carbon/Argon (expands to 75% after Dec 1)
- Carbon/Argon Pricing Event
- Scheduled to double on December 1st
- Benchmark Baseline Rate
- $0.10 input / $0.50 output per 1M tokens (Claude Haiku 5.5, GPT-6 Luna)
- Hydrogen Relative Multiplier
- ~15x cost of Claude Haiku 5.5 / GPT-6 Luna
Gemini 4 Tier Positioning
Google launched Gemini 4 Hydrogen as the high-speed execution counterpart to its flagship Carbon and Argon models. According to Google's guidance, while architectural judgment and complex debugging require Carbon or Argon, Hydrogen is tailored for disciplined tasks: writing implementation code, verifying diffs, and orchestrating test suites. It also replaces prior default models across Google Search AI capabilities and serves as the free tier entry point.
Peculiar Pricing Mechanics
Current pricing for Gemini 4 models creates an unusual transition window:
- Carbon and Argon share identical pricing today, but Google has scheduled both models to double in price on December 1st.
- Argon generates faster responses and consumes fewer tokens per turn than Carbon despite matching raw token rates.
- Hydrogen currently launches with only a 25% discount relative to Carbon and Argon. After the December 1st price hike on the larger models, Hydrogen's effective discount widens to 75%.
- Compared to market entry tiers like Claude Haiku 5.5 and GPT-6 Luna ($0.10 input / $0.50 output per 1M tokens), Hydrogen costs roughly 15x more per token.
Practical Benchmarking
Early testing using Simon Willison's LLM plugin for Gemini shows stark differences in output quality between Hydrogen and the larger models across their supported effort levels, highlighting the trade-offs between pricing, generation speed, and reasoning depth.
When to use
- Writing code implementation and running automated tests in multi-model workflows.
- Everyday free-tier tasks and AI-assisted Google Search workloads.
When NOT to use
- Deep architectural planning requiring Carbon or Argon.
- Workloads requiring low token costs comparable to Claude Haiku 5.5 or GPT-6 Luna.
What to do today
- Audit model usage before December 1st to assess cost impacts of the Carbon and Argon pricing doubling.
- Evaluate Gemini 4 Hydrogen on test orchestration and diff checks before using it in place of Carbon or Argon.
Step names are taken from What to do today.
Step names are taken from What to do today. Audit model usage before December 1st to assess cost impacts of the Carbon and Argon pricing doubling.. Evaluate Gemini 4 Hydrogen on test orchestration and diff checks before using it in place of Carbon or Argon.
- Audit model usage before December 1st to assess cost impacts of the Carbon and Argon pricing doubling.
- Evaluate Gemini 4 Hydrogen on test orchestration and diff checks before using it in place of Carbon or Argon.
