Start with Hydrogen for routine work, try Neon for analysis across several sources or coding tasks that need a plan, and compare Carbon on your hardest multistep problems. Check where you’ll do the work first. The Gemini app, Antigravity, and the Gemini Developer API can offer different models to the same Google account.
Access and prices were checked against Google’s public documentation on November 19, 2026. Google announced the Neon rollout on November 18, so check what your account or project actually shows. The workflows below are suggestions and the costs are estimates calculated from published rates. Google’s Neon announcement
Gemini app: Pick a model for Chat or Tasks
| Plan for a personal Google Account | Chat | Tasks |
|---|---|---|
| No Google AI plan | Auto | Hydrogen |
| Google AI Plus | Auto, Hydrogen | Hydrogen |
| Google AI Pro | Auto, Hydrogen, Neon | Hydrogen, Neon |
| Google AI Ultra | Auto, Hydrogen, Neon, Carbon | Hydrogen, Neon, Carbon |
Google’s plan table covers personal accounts. Remaining usage can also affect availability. Work and school accounts depend on their Workspace license. Gemini app model help
- Organizing meeting notes or rewriting an email. Start with Auto or select Hydrogen directly. Auto routes between Hydrogen and a lightweight model. To use Neon or Carbon, click the current model at the bottom of the prompt box and select it manually.
- Combining several sources. Try Neon on Pro or Ultra. For example, give it interview notes and a requirements document, then ask for a decision table that cites the source of each finding and flags open questions.
- Working through competing constraints. Try Carbon on Ultra. You might use Neon to outline a migration, then ask Carbon to check dependencies, rollback conditions, and missing acceptance criteria.
After selecting a model manually, you can set a Low, Medium, or High thinking level in the same menu. It defaults to Low. Higher settings draw more heavily on usage, so start with the default and raise it if the response misses a key constraint. Gemini app manual model selection
Antigravity: Pick a standalone model or an Orchestrator
| Antigravity plan | Standalone models among these three |
|---|---|
| Free or Google AI Plus | Gemini 3.8 Flash is available. The three models compared here require a higher tier. |
| Google AI Pro | Neon on paid plans or trials with eligible premium devices. Hydrogen on non-trial subscriptions. |
| Google AI Ultra or Gemini Enterprise | Neon, Carbon, Hydrogen |
Open the model menu below the conversation prompt. For a well-defined edit to one file, select Hydrogen directly and specify which test to run. For a migration across several files that needs investigation first, try Neon or select Orchestrator (Neon) so Neon plans and reviews while Hydrogen implements the changes. For a particularly difficult architecture or compatibility problem, try Carbon if your plan includes it and ask for risks and verification steps before making changes. Antigravity model selection
Orchestrator (Neon) is available on Ultra, paid Pro, and Pro trials with eligible premium devices. Orchestrator (Carbon) is available on Ultra. Enterprise lists Neon and Carbon as standalone models, with neither Orchestrator option listed for that plan. Select a standalone model when you want one model to handle every step or want to set a thinking level. The model menu’s View Usage shows your remaining quota. Antigravity Orchestrator documentation
Gemini Developer API: Check project access and estimate cost
Select the project you’ll use in Google AI Studio, check its billing and rate limits, then set the model ID in your application’s request. The Developer API Free tier includes Hydrogen, subject to project eligibility and free usage limits. Neon and Carbon require paid API access. Gemini app subscriptions and API project billing are managed independently. Gemini API billing · Developer API pricing
The example uses paid Standard rates in USD. Assume a set of requests uses 100,000 input tokens and 20,000 billable output tokens, with thinking tokens included in the output count. All input is billed at the uncached rate. Tool charges and taxes are excluded.
| Model ID | Rate period (2026) | Input per 1M tokens | Output per 1M tokens | Example total |
|---|---|---|---|---|
gemini-4-hydrogen |
Published at time of check | $1.50 | $7.50 | $0.30 |
gemini-4-neon |
Published at time of check | $3.00 | $15.00 | $0.60 |
gemini-4-carbon |
Through November 30 | $2.00 | $10.00 | $0.40 |
gemini-4-carbon |
Starting December 1 | $4.00 | $20.00 | $0.80 |
Multiply the input rate by 0.10 and add the output rate multiplied by 0.02. At the same token volume, Carbon costs $0.20 less than Neon through November 30 and $0.20 more from December 1 at the published rates. Models can consume different numbers of thinking and output tokens on the same task. Developer API Standard pricing
Start with Hydrogen on routine requests, such as converting support tickets to JSON with a prescribed set of fields. Pick a few tickets with missing formatting or conflicting details and compare Neon or Carbon using the same inputs and acceptance criteria. Track how often fields pass validation, how long requests take, and how many tokens each uses. Route difficult cases to a different model if the comparison shows a benefit on those cases.
For Carbon-specific access checks and cache rates, see Gemini 4 Carbon access and API pricing. For budgeting across providers, see the AI platform pricing comparison.