AIFast Model Check

AIFast Model Check is a free browser-based diagnostic tool specifically designed to test public HTTPS, OpenAI-compatible API endpoints. By sending fixed probes, it helps developers troubleshoot issues such as inconsistent model routing, capability degradation, or wrapper risks.

Purpose The tool is mainly aimed at developers using API relay stations or third-party interfaces, to screen the performance of models such as Claude, GPT, and DeepSeek on compatible endpoints. It measures interface compatibility, consistency of model declarations, and behavioral sampling results, and can help identify obvious protocol gaps, routing differences, or capability anomalies. However, note that it cannot prove the underlying model identity based on a single black-box test, nor is it a certification from any model vendor.

Main capabilities The tool performs about 3 requests (quick mode) or about 7 requests (standard mode), covering the following checks:

• Authentication and model list: checks Bearer authentication status, the /models return structure, and whether the selected model is visible.
• Protocol fields: checks the HTTP status of Chat Completions, the response JSON, and key fields (such as id, object, created, model, request ID, etc.).
• Token validation: verifies whether input, output, and total tokens are complete, non-negative, and arithmetically consistent, and provides relative token/character fingerprints.
• Dynamic questions: the server randomly generates multi-step calculation parameters and a nonce, and precisely verifies the answer and random values, reducing the impact of fixed-question adaptation.
• SSE streaming output: verifies whether text/event-stream, incremental events, end markers, output concatenation, and streaming usage are identifiable.
• Tool calling: forces a call to a fixed function with a JSON Schema, and verifies whether the tool name, parameter structure, and random nonce are returned as-is.
• Observation signals: fixed prompts collect output style structure, self-reported knowledge cutoff time, etc., and only record observation results without participating in the compatibility score.

How to use

1. Enter a public HTTPS address: for Base URL, enter up to the version path, for example https://api.example.com/v1，不要附加具体接口路径。
2. Use a temporary limited API Key: enter the exact model ID from the console and select "Standard" mode. Detection generates a small number of upstream requests and Token costs.
3. Save the report and repeat the comparison: prioritize checking the response model, Token, dynamic questions, SSE, and tool-call evidence. The total score reflects compatibility and does not equal model identity verification.

The tool provides quick detection (about 3 requests) and standard detection (about 7 requests). Standard detection usually takes 30–120 seconds. The report shows the basis for each item passing, being questionable, or failing. To prevent abuse, the same client is usually limited to at most 6 rounds of detection within 10 minutes, with at most 2 rounds running at the same time.

Applicable scenarios

• Perform risk screening when you suspect a relay station is degrading intelligence or wrapping another model.
• Compare multiple interfaces or evaluate compatibility before migration.
• Troubleshoot specific errors, such as 401, 429, 502, model not found, etc.
• Establish a fixed question set and conduct repeated tests across time periods to identify persistent capability or routing changes.

Limitations

• Detection boundary: The report measures interface compatibility, consistency with model claims, and behavior sampling results. It cannot prove the identity of the underlying model and is not vendor certification.
• Support scope: Only public HTTPS OpenAI Chat Completions-compatible interfaces using a Bearer API Key are supported. Anthropic Messages, Gemini generateContent, OpenAI Responses API, and private custom protocols cannot be used directly.
• Risk of misjudgment: A /models response of 401, 403, 429, or not being public does not mean the model is fake; a single item failing may come from gateway configuration, rate limiting, or differences in model parameters. Judgment must combine multiple pieces of evidence, multiple rounds of detection, and real business question sets.
• Security recommendation: The API Key is used only within this detection request to access the entered relay address and is not written to the database, cache, page address, or detection report. It is recommended to create a temporary Key and revoke it after detection.
• Rate limits: The detection service applies basic rate and concurrency controls, and may temporarily tighten them when busy.