Models, billing, and usage

---
title: Models, billing, and usage
description: Review model prices, balance, request logs, and key limits.
lastUpdated: 2026-07-27
---

## Models and prices

Use the [model catalog](https://app.nexinfer.com/pricing?lang=en) for current public models and prices. The signed-in dashboard applies your account group and permissions. Do not guess model IDs or prices from an old article.

Machine clients should read `pricing_version` from [models.json](https://docs.nexinfer.com/models.json). The public catalog intentionally omits private pricing rules, account groups, and implementation details; request logs remain authoritative for actual charges.

## Conditional charges

- Seedream output up to 2.36 megapixels uses the lower output tier; larger output uses the higher tier. The first reference is included, and each additional reference adds `1/15` of the base image price.
- Seedance settles successful tasks from the model provider's reported `total_tokens` using the configured resolution/input tier. Terminal failed tasks and platform task timeouts refund the pre-consumed quota; a client network timeout alone does not.
- `doubao-tts-2.0` charges by accepted text characters. Invalid speed values are rejected before billing.
- GPT-5.6 cache writes and reads are reported separately. Reusing `prompt_cache_key` does not guarantee a hit unless the prompt prefix is also identical.

## Balance and limits

- **Wallet** shows the current account balance and any available funding options.
- **API Keys** shows each key's status, model scope, and quota settings.
- Key limits and account balance are separate controls. Either can reject a request when exhausted.

A payment method that is not visible in the dashboard should not be treated as available. Use only the funding flow currently presented by the console.

## Review usage

Use **Usage Logs** to filter requests by time, model, key, and request ID. Review:

- success or HTTP error status;
- input, output, and cached tokens;
- cost, time to first token, and total duration;
- model, endpoint, and streaming state.

For a usage discrepancy, record the time and time zone, request ID, model, and a redacted error response before contacting support. Never send the complete key or private prompt.

Models and prices

Use the model catalog for current public models and prices. The signed-in dashboard applies your account group and permissions. Do not guess model IDs or prices from an old article.

Machine clients should read pricing_version from models.json. The public catalog intentionally omits private pricing rules, account groups, and implementation details; request logs remain authoritative for actual charges.

Conditional charges

Seedream output up to 2.36 megapixels uses the lower output tier; larger output uses the higher tier. The first reference is included, and each additional reference adds 1/15 of the base image price.
Seedance settles successful tasks from the model provider’s reported total_tokens using the configured resolution/input tier. Terminal failed tasks and platform task timeouts refund the pre-consumed quota; a client network timeout alone does not.
doubao-tts-2.0 charges by accepted text characters. Invalid speed values are rejected before billing.
GPT-5.6 cache writes and reads are reported separately. Reusing prompt_cache_key does not guarantee a hit unless the prompt prefix is also identical.

Balance and limits

Wallet shows the current account balance and any available funding options.
API Keys shows each key’s status, model scope, and quota settings.
Key limits and account balance are separate controls. Either can reject a request when exhausted.

A payment method that is not visible in the dashboard should not be treated as available. Use only the funding flow currently presented by the console.

Review usage

Use Usage Logs to filter requests by time, model, key, and request ID. Review:

success or HTTP error status;
input, output, and cached tokens;
cost, time to first token, and total duration;
model, endpoint, and streaming state.

For a usage discrepancy, record the time and time zone, request ID, model, and a redacted error response before contacting support. Never send the complete key or private prompt.