Frequently asked questions

For Qwen 3.8 27B, each million tokens costs $0.06 for input, $0.015 for cached input, and $1.00 for output, including reasoning tokens. See current prices.

Top up in the billing dashboard. The minimum is $5, and custom amounts must be in whole dollars. Payments are processed by Polar.

Purchased credit is nonrefundable except where required by law and cannot be transferred between accounts.

Yes. Create named keys in the API keys dashboard to track usage and revoke access separately. All keys share your account balance. Make sure to copy and safely store each key when you create it, since they cannot be viewed again.

Yes. Chat Completions and Responses are supported, including streaming and tool calling. See the integration guide for setup.

Caching automatically reuses matching text at the start of a prompt, which can speed up requests and reduce input costs. System prompts and long conversation history can benefit from caching. For Qwen 3.8 27B, cached input costs 75% less than uncached input.

Qwen 3.8 27B supports a max length of 262,144 tokens across your input and generated output. This includes conversation history, system instructions, and tool definitions.

We are working on adding support for images and videos in the API.

For Qwen 3.8 27B, you can view the official benchmarks here or on Artificial Analysis if you prefer independent evaluation. Scores may not match exactly.

We will work on adding support for Qwen 4 27B shortly after launch, with DeepSeek V4.1 Flash and MiMo V2.6 Flash coming later. Follow our updates page or X account for launch dates and pricing.

No. We do not train models on your API data or retain prompts and responses long term. Account, payment, and usage records are kept to operate the service. See our privacy policy.

The reference speed for Qwen 3.8 27B is 144 output tokens per second once generation starts. Actual speed and the wait before a response begins will depend on prompt length, caching, and demand.

There are no fixed per-account rate limits. Capacity is shared, so requests may queue when the service is busy.

We do not guarantee uninterrupted service. We will publish downtime on our updates page. Contact us if you have a persistent issue.