Send requests to OpenRouter with google/gemini-4-neon. OpenRouter picks the upstream endpoint by default. To pin requests to Google AI Studio, use provider.only to limit eligible endpoints and allow_fallbacks: false to disable failover. The endpoint determines the listed rates and maximum output length.
Suitable tasks. Integrate with OpenRouter, choose between automatic routing and a fixed upstream endpoint, and estimate costs using that endpoint's rates.
Unsuitable tasks. Configure Google's native API or use the listed rates to reconcile charges for past requests.
Applicable model version. google/gemini-4-neon. The model page lists a release date of 2026-11-18.
Applicable clients, agents, or APIs. OpenRouter Chat Completions API.
Recommended reasoning levels and parameters. You can omit reasoning on the first request. The model page lists four levels (low, medium, high, and xhigh). Every endpoint lists Reasoning support, with no breakdown by level.
Store your OpenRouter API key in the OPENROUTER_API_KEY environment variable. The following streaming request pins the model to the Google AI Studio Standard endpoint. The URL and streaming syntax come from the model page's API tab. The provider field comes from the endpoint's Use this endpoint directly example.
curl -N https://openrouter.ai/api/v1/chat/completions \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $OPENROUTER_API_KEY" \
-d '{
"model": "google/gemini-4-neon",
"stream": true,
"messages": [{"role": "user", "content": "Hello"}],
"provider": {
"only": ["google-ai-studio"],
"allow_fallbacks": false
}
}'Omit the entire provider object to use OpenRouter's default routing. With an endpoint pinned and failover disabled, an unavailable endpoint causes the request to return an error. The TypeScript example for the same endpoint enables reasoning with "reasoning": {"enabled": true}.
Send the same short text with default routing and with a pinned endpoint. Check the endpoint actually used, the response, and the usage for each request.
To change routing, start with Providers on the model page. The page describes Balanced as weighing price and speed, Nitro as prioritizing speed, and Floor as prioritizing low prices. Once you choose an endpoint, check its identifier and supported parameters in the endpoint details.
To change the reasoning level or request a long output, test against your chosen endpoint. Supported parameters vary across some Google AI Studio and Google Vertex endpoints, and the endpoint pages list different output limits.
These are the listed endpoint rates as of 2026-11-19, in US dollars per million tokens. All listed endpoints show a context window of 1.05M tokens.
| Upstream endpoint | Included in Standard routing | Input | Output | Cache read | Maximum output |
|---|---|---|---|---|---|
Google AI Studio google-ai-studio | Yes | $3.00 | $15.00 | $0.15 | 65.5K |
Google Vertex Global google-vertex/global | Yes | $3.00 | $15.00 | $0.15 | 1.05M |
Google AI Studio Flex google-ai-studio/flex | No | $1.50 | $7.50 | $0.075 | 1.05M |
Google Vertex US google-vertex/us | No | $3.30 | $16.50 | $0.165 | 65.5K |
Google Vertex EU google-vertex/eu | No | $3.30 | $16.50 | $0.165 | 65.5K |
Google Vertex Priority google-vertex/global/priority | No | $5.40 | $27.00 | $0.27 | 1.05M |
Google AI Studio Priority google-ai-studio/priority | No | $5.40 | $27.00 | $0.27 | 1.05M |
All seven endpoints list parameters for reasoning, tool calling, and structured output. Only the four Google Vertex endpoints list Stop. The three Google AI Studio endpoints omit it.
The $3.00 input and $15.00 output rates at the top of the model page match the listed rates for the two Standard endpoints. The Pricing section also shows the weighted average of what customers actually paid. At the time of collection, the average was $0.2659 for input and $14.78 for output, per million tokens. The page says caching and discounts affect the average.
Routing strategy, an endpoint's Flex or Priority service type, and the model's reasoning level are separate settings. The page lists five of the endpoints in the table outside Standard routing.
The page describes continuation as the way to reach about 1 million output tokens. The chosen endpoint also limits how much output you can get.
Endpoints and rates change over time. For cost estimates, check the chosen endpoint's current listed rates. To verify past charges, check actual usage and billing.
On 2026-11-19, the model page was opened in the international edition of Tabbit to check API, Providers, endpoint details, and Pricing. The sample combines the page's streaming syntax and endpoint-pinning configuration. No OpenRouter API request was sent.
Gemini 4 Neon