Reported by @quartz oyster
Bug Report: GPT-5 returns empty responses with medium reasoning effort despite consuming reasoning tokens
`Steps to Reproduce`
Steps to Reproduce:
- Use GPT-5 model with reasoning_effort="medium" and verbosity="medium"
- Send any text or image analysis request
- Set max_completion_tokens to 1000-4000
- Make API call to OpenAI
`Expected Result`
GPT-5 should return a non-empty response with content analyzing the input, similar to how gpt-5-chat-latest or gpt-4o responds.
`Actual Result`
GPT-5 returns completely empty responses (0 characters) while still consuming 1000-4000 reasoning tokens and taking 17-58 seconds to respond. The API call succeeds (no errors), but content is empty
string. Works fine with reasoning_effort="minimal" or using gpt-5-chat-latest/gpt-4o models
`Environment`
Windows 11 Pro