Skip to content
Back to blog

Gemini 4 Neon vs. Carbon vs. Hydrogen vs. Argon

Compare the four Gemini 4 models by workload and Gemini API terms, then test two models with the same inputs and review criteria.

Published 5 min read
Open Kona Workspace
Turquoise island shallows opening onto an indigo sea at sunset

Start with Gemini 4 Hydrogen for everyday work. Try Neon when a recurring task needs deeper analysis and a quick turnaround. Consider Argon for difficult work where response time still matters, and Carbon for long, demanding jobs with several dependencies or checks. Google released Neon on November 18, adding a model aimed at complex work that teams do frequently. Google’s Gemini 4 Neon announcement

When to use each Gemini 4 model in Kona
ModelWhen to try itExample
HydrogenFrequent tasks with a clear format and straightforward source materialTurn a sales call transcript into next steps and a follow-up email.
NeonRecurring work that needs reasoning across multiple inputsMatch disputed invoices to purchase orders and flag discrepancies.
ArgonComplex analysis that needs close attention to source detailAnalyze how technical requirements and exclusions affect competing vendor bids.
CarbonLong projects with interdependent decisions or calculationsTrace how a policy change affects several linked financial and staffing scenarios.

How to choose by workload

Give Hydrogen a known format and a well-defined job, such as drafting a weekly status update from a project log. Specify which changes count as updates and ask it to flag any open item without an owner. Google designed Hydrogen for rapid execution and offers it on the Gemini API free and paid tiers. Google’s Gemini 4 Hydrogen announcement

Try Neon for a Monday sales review that pulls from a pipeline spreadsheet, account notes, and last week’s forecast. Ask for a table of the largest changes, a source reference for each figure, and a list of discrepancies the team should review. Google positions Neon for complex coding, analysis, and production workflows where speed matters. Google’s Gemini 4 Neon model card

Consider Argon for a vendor renewal where a fee schedule, service exhibit, and amendment each affect what the company owes. Ask for an obligation table and the source clause behind each row. Google describes Argon as an efficient frontier reasoning model that responds faster and uses fewer tokens per turn than Carbon. Google’s Gemini 4 Argon announcement Google’s Argon model pricing and positioning

A board memo is a good test for Carbon. Ask it to reconcile hiring assumptions against a cash forecast, check which figures moved since the previous quarter, and show which conclusions depend on each scenario. Google positions Carbon as its most capable reasoning model for demanding work that takes many steps. Google’s Gemini 4 Carbon announcement

Price and context in the Gemini API

If you are buying API capacity directly from Google, check the date on any cost comparison. Carbon and Argon have introductory per-token rates below Neon’s through November 30, 2026. Their listed rates rise above Neon’s on December 1. Hydrogen has the lowest standard paid rate of these four and is the only one of the four listed on Google’s API free tier. For a job you run often, compare the total cost of producing a usable result, including thinking tokens and any follow-up requests. Google’s Gemini Developer API pricing

Google’s API gives all four models a shared context window of up to about one million tokens across input, thinking, and output. Longer generation requires a specific API configuration. Those figures describe the Google API. Kona lists its own access requirements and subscription information. Google’s long-decode documentation Kona product capabilities Kona pricing

Compare two models on one real task

  1. Pick a completed task with a known answer, such as last month’s expense reconciliation. Give both models the same prompt, files, and template.
  2. Check each result against the original sources. Record missing entries, unsupported claims, arithmetic errors, and minutes spent reviewing or fixing the output.
  3. Decide which result you would actually send. If the faster run takes longer to verify, include that review time in the comparison. For a recurring job, try a second source pack before changing the model your team usually uses.

The decision memo walkthrough gives you a prompt and sample figures you can check by hand. Try the decision memo walkthrough

Kona Workspace offers all four Gemini 4 models in its picker. New conversations use Hydrogen by default unless you’ve saved another preference. Open + (Work options) → Model in the composer. The model control is named Choose model for screen readers. Run each test in a separate conversation. Kona product capabilities Open Kona Workspace