Through an update to its Realtime API, OpenAI is giving developers new components to add to their AI apps. It is also implementing a price reduction when using prompt caching. Finally, OpenAI is adding more speech capabilities to its own chatbot app, ChatGPT.
OpenAI’s Realtime API, whilst still in beta, has received a update introducing five new voices. These additions are specifically designed for voice-centric applications where both input and output are produced through speech interactions.
Cost reduction through prompt caching
The company has implemented a new cost-saving measure through prompt caching. Parts of the prompt that have already recurred in previous prompts are then charged at a lower price. The tool notes repetition because OpenAI uses the same servers for a short period of time to process a user’s API requests.
“This can reduce latency by up to 80% and costs by 50% for long prompts,” the company said. Pricing will be $2.50 per million cached input tokens in text and $20 per million cached input tokens for speech. This approach mirrors that of rival firm Anthropic, which introduced similar cost-reduction measures for voice applications via Claude 3.5 Sonnet last August.
Also read: OpenAI introduces suite of AI tools for developers
ChatGPT gets a voice on the desktop
OpenAI is also increasingly using voice in its own ChatGPT app. The company is now adding the functionality to the app on Mac and Windows. The enhancement delivers more natural-sounding voice interactions and improved handling of paused conversations, building upon previously limited voice features.
Users with a ChatGPT Plus or Team account can now ask for help finding an old chat conversation. The feature is available in the web application by clicking on the magnifying glass in the menu where chat history is listed. Enterprise customers will gain access next week, whilst the feature will be extended to free-tier users the following month.
Tip! Capital or security: OpenAI undergoes consequences of choosing capital