Skip to main content

​
1. Model Usage & Safety

​
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: **feedback@aihubmix.com 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.
GPT-5 is a reasoning model designed for complex inference and structured generation, not for high-frequency real-time tasks. Reasons:
  1. Slower response times due to multiple inference steps.
  2. Higher Token usage (long system prompts and reasoning context).
  3. Translation plugins may accidentally trigger safety policies.
For translation or chat scenarios, use lightweight models like GPT-4o mini or Gemini for faster and more stable responses.

​
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.

​
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 feedback@aihubmix.com 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.
See AiHubMix Privacy Policy for details.

​
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.
Hallucinations are common to all LLMs and do not indicate system failure.

​
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.
Check the API Guide for troubleshooting, or contact support via feedback@aihubmix.com.

​
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.