OpenAI-compat: GPT4V api_base is the full /v1/chat/completions URL, not /v1

Working config (GPT4V, api_base is the full chat URL)

VLMEvalKit GPT4V / OpenAIWrapper posts api_base as-is (requests.post(self.api_base, ...)). A Chat Completions host whose public catalog is GET https://api.pzero.studio/v1/models (no key) fits if that field is the full chat URL.

export OPENAI_API_KEY=<Bearer key for that host>
export OPENAI_API_BASE=https://api.pzero.studio/v1/chat/completions

Or a --config model slot:

{
  "class": "GPT4V",
  "model": "kimi-k2-6",
  "api_base": "https://api.pzero.studio/v1/chat/completions"
}

A /v1 root 404s (the wrapper does not append /chat/completions). Do not use use_azure=True. Do not pass the built-in GPT4o row (hardcoded gpt-4o-*). model is a catalog id with vision. Names containing o1, o3, o4, or gpt-5 swap in max_completion_tokens.

Not asking for a first-party provider. GPT4V plus that api_base is enough.

Start by locating the GPT4V/OpenAIWrapper implementation and its requests.post(self.api_base, ...) call. Compare the documented configuration with the current behavior, then verify the supplied /v1/chat/completions endpoint and model settings; completion requires a clearly defined fix or documentation update.