OpenAI o1 streaming now available + API access for tiers 1–5

Announcements
1

o1-preview and o1-mini now support streaming! You can get responses incrementally as they’re being produced, rather than waiting for the entire response — useful for use-cases that need lower latency, like chat. https://platform.openai.com/docs/api-reference/streaming

We’re also expanding the beta so all developers on tiers 1-5 have access to these models.

15 Likes
4

Awesome! :ok_hand:

For those who now get to use o1 models, and with more interactivity, just a reminder of what it does not accept as input parameters yet…

System Message

  "error": {
    "message": "Unsupported value: 'messages[0].role' does not support 'system' with this model.",

Tools

  "error": {
    "message": "tools is not supported in this model. For a list of supported models, refer to https://platform.openai.com/docs/guides/function-calling/supported-models.",
    "type": "invalid_request_error",

Functions

  "error": {
    "message": "functions is not supported in this model. For a list of supported models, refer to https://platform.openai.com/docs/guides/function-calling/supported-models.",
    "type": "invalid_request_error",

response_format

  • no strict or structured output. Your output desires will likely be lost through all the reasoning.
  "error": {
    "message": "Invalid parameter: 'response_format' of type 'json_schema' is not supported with this model. Learn more about supported models at the Structured Outputs guide: https://platform.openai.com/docs/guides/structured-outputs",

Not max_tokens, but max_completion_tokens

  • and under-specifying can cut the AI off internally before any response is seen, because this now is for limiting your cost of all AI produced language, even the internal whirring.

No temperature, top_p, logprobs, penalties, etc…

  • logit_bias still works - a positive value can make the AI never escape “reasoning”.

Avoid any questions about thinking processes

  • A mere mention of reasoning or internal thoughts will bomb you out.
  • Don’t bother telling it how to think step-by-step.
  • One input, one output is the best usage.

Exception: API Error: {'message': 'Invalid prompt: your prompt was flagged as potentially violating our usage policy. Please try again with a different prompt: https://platform.openai.com/docs/guides/reasoning/advice-on-prompting', 'type': 'invalid_request_error', 'param': None, 'code': 'invalid_prompt'}

(Now you’re talking to o1 – and talking about the pricing, remember, “ChatGPT can make mistakes”)

image
image795×652 49.2 KB

  • Truth: completion_tokens billed as output also includes internally-generated unseen reasoning_tokens.
11 Likes
5

Thanks for reminding! And yep, not *yet — the team’s working on adding system messages, function calling, etc. We’ll post in Announcements when they’re enabled!

5 Likes
6

I know this is not the right forum but I lose nothing… Is streaming also supported in AzureOpen AI too?

I appreciate your understanding.

7

Well, if functions are not working for this model, it is useless.
Any model in the next six months of 2025 that does not have functions available will become obsolete.

1 Like
8

This topic was automatically closed after 11 days. New replies are no longer allowed.