General Questions
A: Nexhina is an OpenAI-compatible gateway, with access to top-tier international models (GPT-4o, Claude Sonnet 4, Gemini, etc.) and more flexible pricing. The interface format is fully compatible, so the OpenAI SDK can be used directly.
A: Any language that supports HTTP can call the API. Python and Node.js have official SDKs for the most convenient experience. For other languages (Go/Java/PHP/Rust), you can use an HTTP client to make requests directly.
A: Contact the administrator to get a test Key, which usually comes with an initial quota.
Call Issues
context_length_exceeded error?A: The input is too long. Streamline the contents of the messages, or switch to a model with a longer context (e.g., gpt-4.1 supports 1M).
model_not_found error?A: The
model parameter is incorrect. Call GET /v1/models to see the list of available models, and pay attention to case sensitivity.A: A network issue caused the SSE connection to drop. There is no way to resume, so you'll need to initiate the request again. We recommend implementing concatenation logic on the client and re-requesting after a stream break.
A: It may be that
max_tokens is set too small, or the model output has reached its limit. Check finish_reason; if it is length, it means the output was truncated. Increase max_tokens.A: Try explicitly asking for a "reply in Chinese" in the system message, or use a model with stronger multilingual capabilities (Claude Sonnet 4 / GPT-4o).
Billing Questions
A: Look at the
usage field in the response. The total number of input + output Tokens is the consumption.A: Set
stream_options: {"include_usage": true}. The last chunk will include usage. Non-streaming requests return usage by default.A: The billing logic is the same (per Token), but the multipliers are different. Nexhina's pricing is more competitive than going direct. See the backend configuration for specific multipliers.
Feature Questions
A: Yes. Models such as GPT-4o / Claude Sonnet 4 / Gemini all support it, and the usage is exactly the same as OpenAI.
A: Yes. Use multimodal models such as gpt-4o / claude-sonnet-4, and pass an image URL or Base64 in
content. See the Multimodal Calling Guide for details.A: Yes. Set
response_format: {"type": "json_object"}. See the JSON Mode Guide for details.A: Not currently supported. You can use the pre-trained models provided by the platform directly, and achieve customization through prompt engineering and few-shot learning.
A: Usually 30 seconds to several minutes, depending on the model and video length. See the Async Tasks Guide for details.
Deployment Questions
A: If the frontend calls the API directly, you'll encounter CORS issues. We recommend going through a backend proxy, or contacting the administrator to configure a CORS whitelist.
A: Nexhina is deployed on the public network, and the internal network needs to be able to access the external network. If it is completely isolated, a private deployment is required.
A: Contact our sales team. We support private deployment in customer data centers.
A: In the admin console → Logs page, you can filter by Key / model / time.
Modified at 2026-06-30 09:28:03