1. Model Usage & Safety
Does AIHubMix store user API request data?
By default, AIHubMix does not store any of the request content you send through our API, nor do we log the responses returned by model providers. AIHubMix acts solely as a proxy, securely forwarding your request to the appropriate model provider and returning their response to you without modification.The only exception
If you proactively report an issue, submit an error ticket, or request assistance with debugging, AIHubMix may temporarily store technical error-related information (such as stack traces, invocation metadata, or endpoint status) to help us diagnose and resolve the problem.These logs do not include your business data, prompt content, or the full request/response payload.
Why do official products like Claude or GPT return different results than the API?
The underlying model is the same; the difference comes from additional engineering optimizations on the web version. Explanation:- The web version is like a fully furnished apartment, with built-in features like search, memory, calculator, and system prompts.
- API calls are like an unfinished apartment, providing only the core capabilities. Developers need to configure context and tools themselves.
Why might using GPT-5 or the “o” series models result in an AiHubMix account suspension?
If you prompt GPT-5 or “o” series models to “show reasoning steps,” “display chain of thought,” or “reasoning trace,” the system may trigger safety policies, which could temporarily restrict or suspend your account. Explanation:- Official safety policies for GPT-5 and “o” series models are stricter; normal use will not trigger a ban.
- If your account is mistakenly flagged or you see abnormal messages, contact support via email: [email protected] for assistance.
- To view model reasoning summaries, use the Response API instead of asking the model directly in the prompt, to avoid triggering safety policies.
How to resolve an account suspension?
If your account has been suspended, you can request reactivation by contacting our online support team or emailing [email protected] . In most cases, suspensions occur when OpenAI’s GPT-5 series models (e.g., gpt-5-nano) are used within immersive translation workflows. These models are designed for advanced reasoning and structured content generation, which makes them unsuitable for high-frequency, real-time translation tasks. Frequent calls in such scenarios may trigger risk control mechanisms, leading to temporary restrictions or suspension of your account. Within OpenAI’s security framework, the GPT-5 series is classified as high-sensitivity models. While normal usage will not result in suspension, repeated use in translation workflows—even without explicitly requesting reasoning outputs—may still be flagged as abnormal activity. To maintain stable account usage, we strongly recommend using non-reasoning models for translation tasks, such as gpt-4.1-mini or gpt-4o-mini, which are better suited for high-frequency requests and carry lower risk of triggering system restrictions. If a suspension appears to be accidental or unusual, please contact support or email us for assistance. However, please note that multiple suspension records may result in permanent deactivation, and recovery may no longer be possible. To minimize risk and ensure continuous access, we recommend avoiding the use of GPT-5 reasoning models in translation workflows.Why is GPT-5 not recommended for translation tools?
GPT-5 is a reasoning model designed for complex inference and structured generation, not for high-frequency real-time tasks. Reasons:- Slower response times due to multiple inference steps.
- Higher Token usage (long system prompts and reasoning context).
- Translation plugins may accidentally trigger safety policies.
Why does GPT-5 sometimes answer “I’m GPT-4” when asked “Who are you”?
This is a known LLM hallucination, where the model inaccurately describes its own foundation, source, or capabilities. Developers using GPT-4, GPT-5, Claude, etc., may encounter confident but incorrect self-identifications. Explanation:- This behavior is not due to platform modifications or output tampering; it’s normal for LLMs.
- GPT-5 was not given the name “GPT-5” during training; the name was assigned afterward by the official release.
- The model does not know its own name or knowledge cutoff; the web version can answer correctly because it has built-in system prompts. Our API version is the official non-web API.
- Asking the model directly via API may produce random or inaccurate answers because it lacks self-awareness.
What Should I Do If Calls to Some Models (like Gemini-3-Pro) Frequently Time Out?
Try increasing the timeout duration. Gemini-3-Pro is a large model, and its inference process often requires a longer reasoning time—especially for complex tasks where the response may take more than 30 seconds. As a result, the default 30-second timeout can easily lead to errors.- If you must use Gemini-3-Pro, be sure to extend the timeout appropriately.
- If fast response time is essential, consider switching to a lighter model such as Gemini 2.0, which works better with shorter timeout settings.
Why did sending just “Hello” consume so many Tokens?
Some third-party tools (like Cline or Claude Code) automatically include context or system prompts in requests, which also count toward Token usage. Even if you only type “Hello,” the backend request may contain extensive chat history or preset text. These extra tokens come from the tool, not the AiHubMix platform.Why do I see 4o-mini usage even though I only called GPT-4o?
Some third-party tools may call lightweight models (like 4o-mini) for conversation summarization, search, or auxiliary computation.Thus, your bill or logs may show multiple models’ Token usage. This extra usage comes from tool configurations, not from AiHubMix automatically switching models.
What is the concurrency limit for API requests?
AiHubMix does not currently impose a uniform concurrency limit. Contact support via [email protected] if you encounter concurrency issues.Why do results vary for the same prompt?
Large language models use probabilistic sampling (e.g., temperature, top-p) to generate text, randomly choosing from multiple possible tokens each time.- Lowering temperature or disabling sampling can make results more consistent.
- Variations may also be affected by context, system prompts, or network conditions.
2. API Calls & Data
Which API endpoints are available?
AiHubMix provides a unified gateway compatible with multiple mainstream model standards:- OpenAI Standard Endpoint:
https://aihubmix.com/v1(supports GPT and compatible models) - Gemini Dedicated Endpoint:
https://aihubmix.com/gemini(compatible with Google native standards) - Claude Auto-Forwarding Endpoint:
https://aihubmix.com(supports Anthropic SDK calls)
What data is recorded during API usage?
We only log necessary usage data: account info, call records, models used, Token consumption, and payment info. Privacy assurance:- User input and model output are not stored.
- Data is used solely for billing and service optimization, not content analysis or third-party sharing.
- AiHubMix does not retain detailed request data; however, underlying cloud providers may log access for security or compliance, governed by their privacy policies.
3. Model Knowledge & Common Phenomena
What is AI Hallucination?
AI hallucination occurs when a large language model generates information that is factually incorrect, unsupported, or entirely fictional. Possible causes:- Biases or gaps in training data.
- Overfitting of model parameters.
- Randomness during generation.
4. Usage & Troubleshooting
How can I monitor API usage and consumption?
You can view call volume, Token usage, and billing details through the AiHubMix dashboard. Supports categorization by model and time period, helping optimize usage and manage costs.What should I do if a call fails or returns an error?
API errors include an error code and explanation. Common causes:- Incorrect request format.
- Model unavailable or usage limit exceeded.
How do I manage my API Key?
Users can generate, revoke, or update API Keys via the dashboard.- Do not expose API Keys in public environments.
- Use separate keys for different projects.
- Rotate keys periodically to ensure account security.