Qwen3.8-27B through an OpenAI-compatible API.
Use a tested 262,144-token context, streaming responses, tool calling, structured output, and reasoning through one familiar API. Prompts and completions are not persisted.
Direct API access is in controlled beta. OpenRouter listing pending approval.
Serving path
Requests pass through the gateway to the inference worker and stream back to you. The gateway does not persist inference content; account, usage, and security metadata is stored separately.
Run Qwen3.8 without replacing your client
OpenAI-compatible API
Use qwen/qwen3.8-27b with existing chat-completions clients. Start with the OpenAI Python SDK guide.
Measured long-context serving
The public path has been exercised through 221K prompt tokens and rejects inputs above its 262,144-token limit. Review the benchmark and conformance results.
Tools, structure, and reasoning
Use function calls, strict JSON-schema responses, and reasoning in streaming or non-streaming requests. See the tool-calling guide.
Flat rates per million tokens
Input (uncached)
$0.24
per 1M prompt tokens
Input (cached)
$0.12
per 1M cached prompt tokens
Output
$2.15
per 1M completion tokens
These configured rates are used to estimate usage charges; payment arrangements are agreed separately. Model details and examples →
Direct beta and OpenRouter distribution
Direct API access is open for requests from developers. Accounts are admitted in small cohorts while capacity and commercial operations remain controlled. Describe your workload.
The same endpoint is pending approval for distribution through OpenRouter.