Gemini 4 Hydrogen vs Carbon vs Argon: Price, Free Tier and Benchmarks

Hydrogen is the only Gemini 4 model on the API free tier

Read the free-tier and app rows first: Hydrogen is the one Gemini 4 model you can try without paying, in AI Studio, the API free tier or the free Gemini app, while Carbon needs Ultra ($99.99 a month and up) or a paid API project. All three share a 1,048,576-token window for input, thinking and output, with a 65,536-token default output that long decode raises to 1 million. Google's sources disagree on Hydrogen in Antigravity: its launch post says paid Google AI Pro and Ultra subscribers, while Antigravity's model table marks it for every plan, free included.

From December 1, Hydrogen costs well under half of Argon or Carbon

For an illustrative month of 50 million input and 5 million output tokens, in ToolColumn arithmetic:

Hydrogen is 25% cheaper now and about 62% cheaper once Argon's and Carbon's introductory rates end. Token counts will not match between models in practice: Google says Carbon spends more thinking tokens per turn than Argon, and each model's thinking level changes how many tokens a task uses. Price each candidate on your own task set before relying on the per-token gap.

Teams on Gemini 3.8 Flash face the opposite timing. Flash costs $0.75 and $3.75 through December 31, 2026 and $1.50 and $7.50 from January 1, 2027, Hydrogen's price, so moving to Hydrogen now doubles the per-token rate for six weeks and costs nothing extra after that.

Google's figures put Hydrogen level with Argon and below Carbon

Read the table as Google's numbers from two posts: Carbon's and Argon's come from the Carbon launch table, run at their highest thinking settings, and Hydrogen's from its own launch post, at the default high level unless marked. On these two rows Hydrogen sits just above Argon and 6 to 8 points below Carbon. Google's terminal-coding numbers do not line up: Hydrogen's 57% is on Terminal-Bench 4.1 (against 18.2% for Gemini 3.8 Flash), while Carbon's 65.2% and Argon's 57.1% are on 4.0, so they cannot be compared.

Antigravity's Orchestrator is Google's way to get both: Carbon or Argon plans and reviews while Hydrogen writes code and runs tests in the same conversation. Google reports that on Terminal-Bench 4.0 it reached near-Carbon success rates while using 24% less quota and finishing 19% faster. Orchestrator (Carbon) is on Google AI Ultra; Orchestrator (Argon) is also on paid Google AI Pro.

Moving from Gemini 3.1 Pro or 3.8 Flash

Google recommends Hydrogen as the upgrade from Gemini 3.8 Flash and older Gemini 3 and 2.5 models. Three API changes need code or settings checks: Hydrogen defaults to high thinking and rejects minimal, it shares one 1M-token window across input and output where 3.8 Flash budgeted them separately, and it supports implicit caching only, so explicit cachedContents calls must go.

Gemini 3.1 Pro is still a preview model at $2 and $12 per million tokens up to 200,000-token prompts and $4 and $18 above, with no free tier. Hydrogen undercuts it on every rate and scores higher than Argon on Google's index, so for work that runs on 3.1 Pro today, Hydrogen is the first model to test; Gemini 4 Argon vs Gemini 3.1 Pro covers moving the harder reasoning work to Argon.