1. Announced new API products and features at DevDay San Francisco.
2. Recordings of keynotes and sessions will be available in a few weeks.
3. Introduced Realtime API for low-latency speech-to-speech experiences.
4. Realtime API is in beta and allows the use of six preset voices.
5. Prompt Caching feature now available, saving up to 50% on cached inputs.
6. Prompt Caching reduces latency by reusing recent input tokens.
7. Model Distillation workflow introduced to train smaller models on larger models.
8. Free fine-tuning tokens available: 1 million per day for GPT-4o, 2 million for GPT-4o mini until October 31.
9. Updated default GPT-4o model reduces input and output token costs significantly.
10. Expanded API access to OpenAI o1-preview and o1-mini.
11. New developer resources including docs, guides, and sample apps added.
12. Promoted upcoming developer gatherings in London and Singapore.
13. Encouraged developers to provide feedback and join the developer forum.
14. Features for generating prompts and structured outputs in the OpenAI Playground.
15. Detailed instructions for using reasoning in routine generation and data generation.
16. Added enhancements to the Chat Completions API for calling functions.
17. Launched beta Evals feature to create custom evaluations on the platform.
18. Improved support for multi-agent systems and structured outputs.
19. DevDay launches and the latest API documentation are accessible online.
20. Introduced visual capabilities for fine-tuning models with text and images.