Overview
Try Gemini 4 Neon if Hydrogen struggles to finish multi-file code changes or reconcile information across several documents. Google released gemini-4-neon on November 18, 2026, on the paid Gemini API tier. Start with Hydrogen for routine tasks such as pulling values from a single workbook. Keep Carbon in the comparison for hard cases where a wrong answer would mean substantial rework. Google's launch announcement
What Changed
- A paid model for complex work. Neon has its own model ID and Standard rates. You can call it through both GenerateContent and the Interactions API.
- A different reasoning default from Carbon and Argon. Neon defaults to
highand supportsxhigh. Carbon and Argon default tomediumand top out athigh. - Another model for Google-managed agents. Google supports Neon in the Antigravity agent in Gemini Managed Agents, either on its own or paired with Hydrogen. Google's developer guide
Compared With Previous Version
Gemini 4 Hydrogen, the preceding general-purpose release, already defaults to high and offers xhigh. It also shares Neon's roughly one-million-token context budget across input, reasoning and output. Neon costs twice as much per uncached input or output token at Standard rates. Test whether it completes enough difficult jobs successfully to justify that premium. Google's developer guide · Gemini API pricing
Google documents the same request schema when switching from Hydrogen, Argon or Carbon. Point a paid test project at gemini-4-neon and set thinking_level explicitly. Start with high on each candidate so you can compare models at the same named setting. Measure completion rate, end-to-end latency and cost per successful task, including retries. Google's migration checklist
Pricing and Access
Standard Gemini API rates checked November 20, 2026. Prices are in USD per million tokens. Output rates include thinking tokens.
| Model | Uncached input | Output | Free API tier |
|---|---|---|---|
| Gemini 4 Hydrogen | $1.50 | $7.50 | Available |
| Gemini 4 Neon | $3.00 | $15.00 | Paid only |
| Gemini 4 Carbon and Argon through Nov 30, 2026 | $2.00 | $10.00 | Paid only |
| Gemini 4 Carbon and Argon from Dec 1, 2026 | $4.00 | $20.00 | Paid only |
Neon's Standard rate for cached input is $0.15 per million tokens. Carbon and Argon have lower input and output rates through November 30. Neon has lower listed rates from December 1 under Google's published schedule. If you run a trial before December, use both sets of rates to estimate the cost of keeping the workload on Carbon or Argon. Google lists Gemini app access through AI Pro and Ultra subscriptions. API calls use project billing. Gemini API pricing · Google's launch announcement
Office File Example
You can pass Word, PowerPoint and Excel files such as .docx, .pptx and .xlsx directly to the Gemini API. Google's document guide demonstrates Office input on Hydrogen, and file slicing works across Gemini 3 and Gemini 4 models. Gemini API release notes · Document processing guide
For a workbook with several tabs, select the relevant worksheet and cell range and ask Hydrogen to extract the figures you need. Keep the same slice when you compare Neon on a harder follow-up, such as reconciling those figures with a second sheet and explaining the discrepancies. Google bills the document input tokens for the selected content sent to the model. Whole-file upload limits and output charges still apply. If your question depends on another sheet, include the relevant range so the model has the figures it needs. Google's file slicing guide
Who Should Upgrade and Who Should Wait
Test Neon for recurring, complex jobs where Hydrogen's errors or retries already cost you time. Use a known-answer workbook or a code change with acceptance tests and compare the cost of getting a result that passes. Stay with Hydrogen when it already meets your quality bar or you rely on free API access. Keep Carbon in the evaluation for difficult debugging and architecture work where the extra reasoning may prevent a costly mistake. These are evaluation starting points based on Google's documentation.




