Call the latest models right
from your Neon backend

Your questions,
answered

• What is AI Gateway?

  AI Gateway is the LLM inference layer built into your Neon project. It runs on Databricks Foundation Model APIs. Use your Neon credential to call models from multiple providers through one endpoint, without setting up separate provider accounts.

• Which models can I call?

  AI Gateway includes frontier and open-weight models from Anthropic, OpenAI, Google, Meta, Alibaba, Zhipu AI, Moonshot AI, Thinking Machines, and others. The catalog changes as models are added or retired, and availability can vary by region. Check the live model catalog for current models, prices, and supported endpoints.

• What is the difference between Neon AI Gateway and Databricks Unity AI Gateway?

  Both use Databricks Foundation Model APIs for model serving. Neon AI Gateway is built into Neon projects and credentials for developers building applications and agents. Databricks Unity AI Gateway provides centralized governance for AI traffic within the Databricks Platform.

• How does AI Gateway relate to the rest of the Neon backend?

  AI Gateway shares the same project and branch boundaries as Lakebase Postgres, Functions, Object Storage, and Managed Better Auth. A Function receives NEON_AI_GATEWAY_TOKEN and NEON_AI_GATEWAY_BASE_URL alongside DATABASE_URL, so it can call a model and write results to Postgres without separate provider keys.

• Do I need to run my application on Neon to use AI Gateway?

  No. Any application or service that can make an HTTPS request with a bearer token can call AI Gateway. Running model calls in Functions keeps them close to the other services in your Neon backend and lets them follow the same branching workflow.

• What happens to AI Gateway when I create a branch?

  Each Neon branch has its own AI Gateway endpoint. Requests from a development or preview branch stay scoped to that branch, separate from production.

• Do I have to change my code to use AI Gateway?

  For an OpenAI-compatible client, change the base URL and credential. The request and streaming formats stay the same. AI Gateway also provides provider-specific endpoints when you need features from the OpenAI Responses API, Anthropic Messages API, or Gemini API.

• Who can use AI Gateway after GA?

  AI Gateway is available on the Neon Launch and Scale plans. Both plans have the same AI Gateway pricing. You need prepaid AI Gateway credits before you can make inference requests. Any paid account with prepaid credits can use every model in the catalog, subject to regional availability.

• How does AI Gateway pricing work?

  Inference is billed per token at the published rate for each model, with no additional Neon markup. One AI Gateway credit equals $1 USD. The minimum credit purchase is $5, and purchased credits are valid for 12 months. Check the model catalog for current per-model rates.

• What are the rate limits?

  AI Gateway applies a default limit of 200,000 tokens per minute (TPM). These are default limits and can be increased on request. If you hit a limit, requests return HTTP 429 and the error body explains which limit you reached. Contact support if you need a higher TPM limit.

• How do I buy and manage credits?

  Buy credits from the Billing page in the Neon Console. Credits are added to your AI Gateway balance after payment and accumulate across purchases. You can see your current balance in the Neon Console. Automatic top-ups can add credits when your balance falls below a threshold.

• How are credits deducted?

  Each inference request deducts credits based on the model's token price and the final number of tokens reported by the serving system. Balance updates can take about five minutes. A failed request can still be billed if the model consumed tokens before the request failed.

• What happens when my credit balance runs out?

  AI Gateway stops accepting new inference requests when the available balance is too low. Requests already in progress can finish and are charged for the tokens they consume. Because usage reporting is not instantaneous, your balance can fall slightly below zero. Neon applies a $2 minimum balance to account for this delay.

• Do credits expire?

  Purchased credits are valid for 12 months from the purchase date. Promotional credits can have a different expiration date. Check the Billing page or the terms of the promotion for the applicable date.

• Why can't I access a model in the catalog?

  Model availability can vary by region, and the catalog changes as models are added or retired. Use the authenticated GET /v1/models endpoint to see the models available to you.