Models and pricing

Discover supported models

curl https://agents.naven.network/ai-gateway/v1/models

This free endpoint returns the complete Vercel catalog compatible with Naven's text completion contract. Use the returned id as the request's model. The gateway refreshes the upstream catalog automatically, so model IDs, availability, and prices can change without a Naven release. Read the catalog instead of hardcoding a list or a price.

Jev is available through the separate Evaluations endpoint. It is an evaluation model rather than a language model, so it appears in the response's evaluation object and not in the text-model data array.

In the playground, open Choose a model to browse the entire compatible catalog in a dialog. Search by model name, model ID, or provider, or use the provider filters. Gemini 2.5 Flash Lite, GPT-4.1 mini, and Claude Haiku 4.5 are marked as recommended starting points when available; they are not the complete list.

“Complete” means every Vercel model that can honor this endpoint's bounded text Chat Completions request and a deterministic price. The catalog includes language models with text input and output, a supported output limit, sufficient context capacity, and published pricing that this service can charge safely. Image, video, audio, embedding, reranking, realtime, and other model types are not part of this endpoint. Models with provider-dependent pricing, unsupported capabilities, or reasoning that cannot be turned off are also excluded. Optional reasoning is disabled for this version of the gateway. A model's inclusion does not guarantee provider uptime for every call.

Use the model picker

1. Open the text-generation playground.
2. Select the current model row to open Choose a model.
3. Search or filter by provider. Each result shows its live flat price in USDG.
4. Select a model, enter your prompt, and review the x402 payment offer before signing.

The count in the picker comes from GET /models. If the live catalog cannot be loaded, the dialog shows an error and retry action instead of presenting the three recommendations as the complete list. Paid calls remain disabled until current prices are available.

Price per successful call

Each model has a flat USDG price for one completion within the gateway's limits. The price includes the service fee. It is a request package, not a bill calculated from the response's actual token usage. A shorter prompt, a smaller max_tokens, or an early completion does not reduce the charge.

priceUsd is the human-readable USDG amount. amount is its integer value in the token's smallest unit: 1 USDG = 1,000,000 atomic units. Compare and store amount as an integer string or BigInt rather than a floating-point number.

Catalog prices are refreshed periodically. Send the intended request to POST /chat/completions without payment to receive the current price as a standard HTTP 402 payment requirement. Review the returned amount and recipient before signing. The paid retry must still match the current model price, token, network, and recipient.

Payment authorizations are valid for a short period. If an unsigned offer becomes stale, request a new one. If you already sent a signed request and lost its response, follow the recovery instructions before creating another payment.

Request limits

Streaming, images, audio, embeddings, tools, structured-output parameters, custom reasoning settings, provider overrides, and other request fields are not accepted. Model context windows may be larger than the gateway's limits; the limits above still apply.

Payment network

Generation failure does not request settlement. If payment confirmation is uncertain after generation, keep the original signed request and use the recovery guide.