Transcript of OpenAI DevDay: Keynote Recap
OpenAI
Key takeaways
- GPT-4 Turbo can handle up to 128,000 tokens of context and includes a JSON mode that guarantees valid JSON output.
- The platform now offers built-in retrieval, allowing developers to incorporate external documents or databases into their applications.
- OpenAI introduced a Custom Models program where researchers collaborate with companies to build tailored models for specific use cases.
- Rate limits for established GPT-4 customers have been doubled, with options to request further increases directly in API settings.
- GPT-4 Turbo is priced approximately three times cheaper for prompt tokens and twice cheaper for completion tokens compared to GPT-4.
- GPTs are customizable ChatGPT instances that can be created without coding, shared privately or publicly, and will be available through a forthcoming GPT store.
- The new Assistance API adds persistent conversation threads, built‑in retrieval, a sandboxed Python code interpreter, and enhanced function calling.
