AI summary by TwinMind
OpenAI’s new API is 200x more expensive than competition
Theo - t3․gg
26m 2s
OpenAI O1 Pro API Release and Pricing Shock
- O1 Pro API launched at $150/million input tokens, $600/million output tokens
- 136x more expensive than O3 Mini (best current model)
- 200x more expensive than DeepSeek R1 reasoning model
- OpenAI lost money on $200/month ChatGPT Pro subscribers using O1 Pro heavily
- Model represents massive departure from industry trend toward cheaper AI
Industry Pricing Trends and Competition
- AI model costs have been racing downward consistently
- Gemini 2.0 Flash: comparable quality to GPT-4, 20x cheaper, faster
- Google's advantage: own processors, architecture, data, scientists, infrastructure
- DeepSeek V3: originally priced close to Gemini, later increased 2x input/3-4x output
- DeepSeek R1: powerful open reasoning model, relatively cheap
- Mistral 3.1: launched same week as O1 Pro, undercutting Gemini 2.0 Flash
Model Performance Comparisons
- O3 Mini consistently outperforms O1 Pro in testing
- O3 Mini: faster, cheaper, better answers than O1 Pro
- GPT-4.5: massive model with more parameters, feels more human but not 7-20x better
- O3 Mini High: best current model for most use cases
- Claude 3.7: more expensive than 3.5, generates more tokens, sometimes worse code answers
Technical Issues with O1 Pro
- Frequent failures when switching tabs or refreshing during generation
- Mobile app: 50%+ failure rate when switching apps
- No way to copy messages from ChatGPT interface
- UI bugs: random element selection, missing copy buttons, poor scrolling
- Takes 2+ minutes for simple problems, often produces wrong answers
- Can't resume interrupted generations (unlike Grok)
Token Economics and Reasoning Costs
- Tokens: 4-8 character chunks used for AI processing
- Reasoning models generate hidden "thinking" tokens before answers
- Users pay for all tokens including hidden reasoning steps
- O3 Mini High uses more tokens due to extended reasoning time
- Low/Medium/High settings control reasoning effort and token generation
Practical Implications for Developers
- T3 Chat: built due to frustration with existing AI interfaces
- O1 Pro unlikely to be added to T3 Chat due to cost/performance ratio
- GPT-4.5: only ~30 messages generated since launch with bring-your-own-key
- O3 Mini High may need to move to premium tier due to high token usage costs
- Durable objects could solve connection interruption issues but implementation is complex
Market Analysis and Future Predictions
- O1 Pro release seems like response to "racing to the bottom" trend
- OpenAI appears scared of not shipping years-long development projects
- Models may be foundational for future O4, O5, O6 development
- Industry focus remains on better value models
- O3 Mini represents step toward best options being best priced
