Power Digital

// AI News

Gemini 4 Neon vs Hydrogen: prices and which to use for 5 jobs

Alex Ng
AI News
Illustration with the headline: Gemini 4 Neon Vs Hydrogen

In short: Gemini 4 Neon costs $3 per million input tokens and $15 per million output tokens, exactly double Gemini 4 Hydrogen at $1.50 and $7.50. For most routine business jobs Hydrogen is the cheaper starting point, and Neon earns its price on long, tool-heavy, high-stakes work.

Google published both models on its own blog: Gemini 4 Neon (18 November 2026) and Gemini 4 Hydrogen (12 November 2026). Below are the facts, a price table in US dollars and Singapore dollars, and a guide to which model suits five common business jobs.

What happened

Google released Gemini 4 Hydrogen on 12 November 2026 and Gemini 4 Neon on 18 November 2026. Both are generally available according to the Google AI Studio changelog. They sit in the same Gemini 4 family, which also includes Carbon and Argon.

Google places Hydrogen as the fast, everyday execution model, which it calls a "daily driver for agentic workflows". Neon sits between Hydrogen and Carbon. Google says Carbon handles the hardest long-horizon projects, and Neon balances depth against efficiency. Google also says Neon can act as an orchestrator that directs several Hydrogen sub-agents in Antigravity, its agent environment.

Hydrogen is the default model for the Antigravity agent in Gemini Managed Agents, according to the changelog.

Key details

Price table

All prices are per one million tokens, taken from Google's pages. SGD figures use the USD to SGD rate of 1.2896, the European Central Bank reference rate for 18 November 2026 as published by the Frankfurter rate service (the latest available on 19 November 2026). SGD values are rounded and exclude any GST or card fees.

ItemGemini 4 Hydrogen (USD)Gemini 4 Hydrogen (SGD)Gemini 4 Neon (USD)Gemini 4 Neon (SGD)
Input$1.50S$1.93$3.00S$3.87
Cached inputNot publishedNot published$0.15S$0.19
Output$7.50S$9.67$15.00S$19.34

The Hydrogen price is the same across its whole 1 million token context window.

Specifications and availability

ItemGemini 4 HydrogenGemini 4 Neon
Release date12 November 202618 November 2026
Model IDgemini-4-hydrogengemini-4-neon
Context window1 million input tokens1 million tokens
Output limitUp to 1 million output tokensLong decode up to 1 million tokens
Thinking levelsLow, medium, high (default), xhigh (new)Four levels, default high, top level xhigh
Free API tierYes, per the changelogPrices shown are for the paid tier

Both models are reachable through the Gemini API and Google AI Studio, Android Studio, Gemini for business, Google Antigravity (paid Google AI Pro and Ultra plans) and the Gemini app model dropdown. Hydrogen also reaches free Gemini app users through automatic model selection, appears in Google Search for all users, and is rolling into Gemini in Google Sheets. Google has not announced Singapore availability for either model.

Google's own benchmark claims

These figures are Google's own claims.

  • Neon: Artificial Analysis Intelligence Index of 61 at xhigh, ranked third overall; first on the Artificial Analysis Coding Agent Index at 58.6; 65.2% on Terminal-Bench 4.1.
  • Hydrogen: Intelligence Index of 51 at high and 54 at xhigh; 57% on Terminal-Bench 4.1, against 18.2% for Gemini 3.8 Flash.
  • Cost: Google says Neon at high thinking costs 32% less per task than Carbon. Hydrogen is said to finish tasks 2.5 times faster than Gemini 3.8 Flash at its default setting.

How it compares

The price gap is simple: Neon is exactly 2 times Hydrogen on input and on output. The cost per job is not always 2 times, because a more capable model may need fewer retries or fewer tokens. Google has not published a Neon against Hydrogen cost-per-task comparison.

On Google's own numbers, the Artificial Analysis Intelligence Index gap is 61 (Neon, xhigh) against 54 (Hydrogen, xhigh), and 57% against 65.2% on Terminal-Bench 4.1. Hydrogen is not a lightweight model by those scores, but Neon is ahead.

Cost per 1,000 tasks

The token counts below are assumptions, and real usage depends on prompts, tool calls and how much the model thinks. Google has not published how thinking tokens are billed. Cost = (input tokens x input price + output tokens x output price) x 1,000 tasks, with no caching.

Job (assumed tokens in / out per task)Hydrogen, 1,000 tasksNeon, 1,000 tasks
Customer chat reply (4,000 / 400)$9.00 (S$11.61)$18.00 (S$23.21)
Summarise a long PDF (40,000 / 1,000)$67.50 (S$87.05)$135.00 (S$174.10)
Spreadsheet reconciliation (60,000 / 3,000)$112.50 (S$145.08)$225.00 (S$290.16)
Draft a proposal (8,000 / 3,000)$34.50 (S$44.49)$69.00 (S$88.98)
Coding agent task (200,000 / 20,000, whole session)$450.00 (S$580.32)$900.00 (S$1,160.64)

A business that handles 5,000 customer chats a month would pay about S$58 on Hydrogen and S$116 on Neon at these assumptions. At that scale the price difference is small next to staff time, so the real question is whether the answers are good enough.

What it means for businesses

Which one for which job

This table is Power Digital's view, reasoned from Google's prices and claims.

Business jobStart withWhy
Customer service chat and FAQ answersHydrogenShort tasks, high volume, and cost scales with volume. Hydrogen also has a free API tier for prototyping. Move up only if your own test set shows wrong answers.
Summarising contracts, PDFs and meeting recordingsHydrogen, then test NeonBoth take a 1 million token context. Neon is stronger on Google's legal research and tax benchmarks, so test it on documents where a missed clause is costly.
Spreadsheet, finance and accounting workNeonGoogle reports Neon first on most finance and spreadsheet benchmarks from Surge AI. Errors here cost money, so the extra price is easier to justify. Verify on your own sheets.
Drafting proposals and reportsHydrogenOutput volume is modest and a person edits the result. Neon's extra depth matters less when a human reviews it.
Coding and multi-step agents with many tool callsNeon for the planner, Hydrogen for the workersGoogle describes exactly this split in Antigravity. It keeps most tokens on the cheaper model.

Mistakes to avoid

  • Choosing by leaderboard rank alone. A rank on a public index says little about your invoices, your product catalogue or your customers' wording.
  • Sending every request to Neon. If most of your traffic is simple, routing it to Hydrogen halves that part of the bill at these list prices.
  • Ignoring the thinking level. Both models offer four levels, and higher levels usually mean more tokens and more waiting. Set the level per job instead of leaving one setting for everything.
  • Skipping a review step. Google's safety figures are its own internal evaluations, so keep a human check on anything that touches money, contracts or customer promises.

A practical way to decide

  • Collect 30 to 50 real examples from your business, including a few hard ones.
  • Run them on Hydrogen first and record how many a staff member would accept without change.
  • Run the failures on Neon. If Neon fixes most of them, route only that type of job to Neon.
  • Check where data is processed and under which terms before sending customer data, particularly personal data covered by Singapore's PDPA.

If you want a custom agent that routes between models like this, our AI agent development service covers that kind of build. To estimate your own bill, use the token cost calculator, and if you are weighing other vendors, read which Claude model to pick for a similar tier decision.

Our take

This is Power Digital's view. The price structure makes the choice easy to reason about: Hydrogen is the default and Neon is a deliberate upgrade for specific jobs. Most Singapore small businesses will not feel the price gap on chat or drafting work, so quality on your own data should decide, not the benchmark ranking.

Rate limits for either model have not been published, and neither have Singapore-specific data-handling terms. Settle both before you commit a production workload.

Sources

Tags

#Google AI #Model comparison #Gemini 4 #AI pricing

// what we build

Explore Our Services

// AI News

Insights & Resources

Read more →
Back to Articles