ModelBrief · Gemini 4 Neon · Topic Page
Last updated Nov 20, 2026

Gemini 4 Neon vs Hydrogen: What 2x the Price Buys

Published Nov 19, 2026 · Last updated Nov 19, 2026

In the Gemini 4 family, Hydrogen is the fast tier and Neon is the mid-tier, and the price differs by exactly 2x: Neon at $3/$15, Hydrogen at $1.50/$7.50. Both share a 1M context, four thinking tiers and the same input/output modalities. So what does twice the price actually buy? The answer: a 7-point AA index gap (61 vs 54), No. 1 finishes in coding and enterprise knowledge work, and a higher bar to entry — Neon has no free tier. Which one to use day to day depends on whether your work is hard enough to justify paying double for 7 points.

Price: exactly 2x, plus the free-tier gap

ItemNeonHydrogen
Standard (input / output, per 1M tokens)$3 / $15$1.50 / $7.50
Cached input$0.15$0.075
Batch / Flex$1.50 / $7.50$0.75 / $3.75
Priority$5.40 / $27.00$2.70 / $13.50
Free tierNoYes

Every tier is exactly 2x. Two equivalences are worth remembering: Neon's Batch / Flex price ($1.50/$7.50) equals Hydrogen's standard price; Hydrogen's Priority price ($2.70/$13.50) is still below Neon's standard price. The free tier is the biggest barrier difference: Hydrogen's API has a free tier and free users can use it in the Gemini app; Neon's API is paid-tier only and the Gemini app needs a Pro / Ultra subscription. For zero-budget trials and light calls, Hydrogen is the only option.

Scores: 61 vs 54, and where the gap sits

On the AA index v5.0.1: Neon at 61 (xhigh) and 60 (default high); Hydrogen at 54 (xhigh) and 51 (default tier). A 7-point gap at the top tier, 9 points at the default tier. That is the second-largest gap inside the family (Carbon vs Hydrogen is 60 vs 54, a 6-point gap). Neon's lead concentrates in coding and enterprise knowledge work: the AA Coding Agent Index at 58.6, No. 1; the Vals Index at 70.9%, No. 2; AutomationBench at 56.3% (Google-reported). Hydrogen's scores on those individual tests were not verified one by one in this round, so no direct comparison there — but a 7–9 point composite gap already says Neon handles complex multi-step work a tier better.

One caution: Neon's 61 needs the xhigh tier, and one xhigh index run generates 283M output tokens — 2.3x the high tier's; Hydrogen's 54 is likewise its xhigh figure. Both top-tier scores are "most expensive tier" scores; the default-tier gap (60 vs 51) is the more honest comparison.

Thinking tiers: same four, same default

Both offer low / medium / high / xhigh, both default to high. The difference is not the number of tiers but their quality: Neon is adapted from Carbon, while Hydrogen is the family's fast base (succeeding 3.8 Flash). Google's design intent is clear: four tiers let users switch models under the same control scheme — moving from Hydrogen to Neon needs no new tier vocabulary, just twice the price for higher tier scores.

Speed: no head-to-head measurement exists

Google positions Hydrogen as "fast everyday execution" and Neon as "fast end-to-end task completion", but the launch materials contain no same-workload timing comparison between the two. The only known figure is Neon's: AA measured 21.5 minutes per index task and 56.51 seconds to first answer token at xhigh — among the slowest in its price tier. Hydrogen's same-measure timing was not verified in this round, so no guess here. The only fair statement: Neon's "fast" is relative to Carbon; relative to Hydrogen it is the more expensive, more capable and slower one — "fast" for high-frequency complex work means finishing jobs cleanly, not delivering the first token sooner.

Subscriptions and availability

In the Gemini app, Hydrogen is available to free, Pro and Ultra users ("Everyday help" is 4 Hydrogen); Neon needs a Pro / Ultra subscription ("Deeper analysis"). On the API, Hydrogen has a free tier and Neon does not. In Antigravity both need paid Pro / Ultra, and Neon can also act as an Orchestrator directing Hydrogen subagents — the combination Google designed: Neon judges, Hydrogen executes.

Decision table

Task typePickWhy
Everyday coding, interactive programming, quick loopsHydrogenHalf the price, has a free tier; Neon's 7-point edge does not show on these tasks
Zero-budget trials, light callsHydrogenNeon has no free tier
Complex multi-step coding agents, enterprise knowledge workNeonAA 61 vs 54; Coding Agent Index No. 1; Vals 70.9%, No. 2
Long-horizon agent work, tough debuggingNeon (or Carbon)Hydrogen's capability ceiling is lower; the hardest jobs are covered in our Neon vs Carbon comparison
Non-urgent batch workNeon BatchBatch at $1.50/$7.50 equals Hydrogen's standard price, with a tier more capability
Free Gemini app usersHydrogenNeon needs a Pro / Ultra subscription
Antigravity orchestrationNeon + Hydrogen togetherNeon as Orchestrator, Hydrogen as subagents — Google's official combination

Migration notes

Moving from Hydrogen to Neon means changing only the model ID (gemini-4-hydrogen → gemini-4-neon); the request schema is unchanged. Three things to watch: first, the bill doubles — the same call volume costs exactly 2x, so validate the effect on small traffic before switching everything; second, the free tier disappears — calls that used to run free now need a paid tier and a billing account; third, the tier vocabulary is the same but the scores differ — at high, Neon scores 60 and Hydrogen 51, so do not carry old tier habits over unchanged.

Our take: Neon's 2x price buys a 7–9 point AA index gap and No. 1 finishes in coding and enterprise knowledge work, at the cost of losing the free tier. For fast everyday execution, interactive coding and zero-budget trials, Hydrogen remains the right answer; for high-frequency complex coding and enterprise knowledge work, Neon's double price is money well spent. The cheapest way to use both is not to choose at all but to combine them: Neon as Orchestrator judging, Hydrogen as subagents executing — that is the division of labor Google designed.

Reference sources: this article is an original rewrite produced by this site; facts and figures were checked against Gemini API pricing (read Nov 19, 2026), Artificial Analysis' model page (read Nov 19, 2026) and Google's official blog (Nov 18, 2026)

← Back to the Gemini 4 Neon topic page