OpenAI’s Most Expensive Model Yet!

Key Concepts

• 01 Pro: OpenAI's new, highly expensive reasoning model.
• Reasoning capabilities: The model's ability to perform complex logical inferences.
• Token pricing: The cost per million input and output tokens.
• Context window: The amount of text the model can process at once (200,000 tokens for 01 Pro).
• API endpoints: Specific URLs used to access the model's functionality (responses API).
• Batch processing: Processing multiple requests at once, suitable for large-scale tasks.
• Function calling: The model's ability to use external tools or functions.
• Structured output: The model's ability to generate data in a specific format (e.g., JSON).
• Agentic workflows: Systems where AI agents autonomously perform tasks.
• Rate limits: The maximum number of requests allowed within a given time period.
• SVG generation: Creating Scalable Vector Graphics using the model.
• Chain of Thought: The model's internal reasoning process.

01 Pro Overview and Pricing

• OpenAI has released 01 Pro, a new model accessible via API.
• The primary distinguishing factor is its high price: $150 per million input tokens and $600 per million output tokens.
• This is twice as expensive as GPT-4.5 and 1000 times more expensive than GPT-4o mini.
• The model is likely targeted towards large institutions and research, not general developers.
• Example: Creating a pelican SVG costs almost a dollar.

Feature Comparison

• Reasoning: 01 Pro is positioned as the best model for reasoning.
• Speed: It is slower than GPT-4.5 and GPT-4o.
• Input: Accepts both text and images, similar to GPT-4.5 and GPT-4o.
• Output: Generates text only.
• Context Window: 200,000 tokens, with a maximum output of 100,000 tokens.
• Cut-off Date: Same for all three models.

API Limitations

• Does not support the chat completion endpoint, requiring custom implementation for chat applications.
• No real-time capabilities or streaming support.
• Available only through the responses API.
• Supports batch processing, function calling, and structured output.
• Cannot be fine-tuned.

Access and Rate Limits

• Available to all paid tiers (Tier 1 to Tier 5).
• Rate limits vary by tier, similar to GPT-4o.
• The price is approximately 10x more expensive than o1 and 1000x more expensive than GPT-4o mini.

Simon Wilson's SVG Generation Test

• Simon Wilson tested 01 Pro by asking it to generate an SVG of a pelican riding a bicycle.
• Initial attempt cost about 94 cents (close to a dollar) with 1,500 output tokens, but the output quality was not great.
• Setting the "effort" parameter to "high" resulted in a better image but used more reasoning tokens (960 vs. 104).
• The Chain of Thought is not accessible, which is a drawback given the high cost.

Market Segmentation

• OpenAI seems to be targeting two market segments:
  • Developers building applications for their own customers (suitable for models like O3 mini or a mini version of reasoning models).
  • Large institutions using high-reasoning models for internal research and usage.
• Example: Grab saw a 6% improvement in accuracy on complex prompts using 01 Pro compared to 03 mini High.

Pricing Trends

• Sam Altman has stated that there is a 10x reduction in pricing for the same level of intelligence over a year.
• Hopes for a similar trend with GPT-4o, GPT-4.5, and 01 Pro.

Conclusion

01 Pro is a powerful but expensive reasoning model aimed at specific use cases requiring high accuracy and complex problem-solving. Its high cost and API limitations make it unsuitable for general developers, but it may be valuable for large organizations and research institutions. The hope is that prices will decrease over time, following historical trends.