Skip to content

0 | FAQ

ChatGPT Plus is Rate limited:
50 msgs/week for o1-preview
50 msgs/day for o1-mini
Context Window is 128000 tokens
Max Tokens in Completion Output is 32768 (o1-preview) - 65536 (o1-mini)

Stats

Read the o1 System Card

  1. Scorecard - with safety evaluations and a Preparedness Framework
  2. Report - 42 page PDF with detailed information on the model data and training, evaluations and more.
1. What is o1 and how is it different?

Is a new series of reasoning models that are designed to solve hard problems by thinking before they respond.

Uses reinforcement learning to hone its chain-of-thought, refine strategies it uses, recognize & correct mistakes.

See: Learning to Reason with LLMs to understand how the model learns to reason and how this breaks down tasks compared to gpt-4o.

2. How do I get access to o1?

The model family is currently available from OpenAI, Azure OpenAI and GitHub Models Marketplace. However access for developer usage is limited and requires additional steps and approvals which makes it hard to use for casual exploration. Currently the best option is to use ChatGPT Plus (paid subscription) but note that rate limits keep evolving for both API-based and Subscription-based access.

  1. GitHub / OpenAI o1-mini - Join the waitlist
  2. GitHub / OpenAI o1-preview - Join the waitlist
  3. Azure OpenAI / o1-preview - Request access
  4. Azure OpenAI / o1-preview - Request access
  5. OpenAI / API Access - Only Usage Tier 3,4,5 have access, with tier-specific rate limits
  6. OpenAI / ChatGPT Plus - 50 msg/week (o1-preview) and 50 msg/day (o1-mini)
3. What are the LIMITATIONS?

These are areas where gpt-4o has a natural advantage for now, since it supports them

  • Text generation only (no images for now)
  • User & assistant messages only (no system messages)
  • Streaming not supported (wait till done)
  • No function calling, response format parameters (tools)
  • Fixed parameters (temp=top_p=n=1, presence_penalty=frequency_penalty=0)
  • Models not support for Assistants API or Batch API
4. What are the CAPABILITIES?

  1. Who is it for? - Researchers, educators, developers building complex workflows. Anywhere that we have a hard problem that needs to be solved with multi-step reasoning.

  2. o1-mini - faster and 80% cheaper haan o1-preview but its world knowledge is not as broad

5. What are REASONING TOKENS?

New concept introduced in o1. Tokens used primarily for the thinking steps - once it produces the answer (visible completion tokens), it discards reasoning tokens from context. This is what a multi-step conversation looks like now.

Reasoning Image

6. What are some PROMPTING TIPS?

From the docs

These models perform best with straightforward prompts
Some prompt engineering techniques, like few-shot prompting or instructing the model to "think step by step," may not enhance performance and can sometimes hinder it. Here are some best practices:

  • Keep prompts simple and direct:
    The models excel at understanding and responding to brief, clear instructions without the need for extensive guidance.
  • Avoid chain-of-thought prompts:
    Since these models perform reasoning internally, prompting them to "think step by step" or "explain your reasoning" is unnecessary.
  • Use delimiters for clarity:
    Use delimiters like triple quotation marks, XML tags, or section titles to clearly indicate distinct parts of the input, helping the model interpret different sections appropriately.
  • Limit additional context in retrieval-augmented generation (RAG):
    When providing additional context or documents, include only the most relevant information to prevent the model from overcomplicating its response