max_completion_tokens is not affective

[hopewise]

Describe the bug
I am trying to set max tokens received while using assistant API, I tried to max_completion_tokens: 500, as:

run_response = client.runs.create(
          thread_id: thread_id,
          parameters: {
            max_completion_tokens: 500,
            assistant_id: ENV['OPENAI_ASSISTANT_ID'],                      
            additional_instructions: system_message + "\n\n",                
            stream: proc do |chunk, _bytesize|
            ...

Expected behavior
Precise response.

Actual behavior
Too long response.