rewrite this title OpenAI Unveils Its Most Advanced Conversational Model: gpt-realtime

OpenAI has introduced gpt-realtime, a new conversational model that better follows complex instructions, produces more natural-sounding voices, and is 20% cheaper. The model comes with two new voices.

In October 2024, OpenAI introduced the Realtime API, which allows developers to create low-latency, multimodal experiences. Since then, thousands of developers have used this API to build natural-language conversation experiences into their applications. Now, a more advanced conversational model, gpt-realtime, has been announced, which includes enhanced features.

More Advanced and Cheaper

The new model is better at following complex instructions, and the error rate in tasks like hailing a ride has been reduced. The voices generated by the model are also noted to be more natural and expressive. OpenAI states that the model can interpret system messages and developer commands much better than previous models.

The initial launch of the Realtime API offered six voice options, with two new voices added later. Now, two more new voices named Marin and Cedar have been announced. Additionally, the eight existing voices have been updated to provide a more natural and fluid experience.

The new model also performs impressively in performance tests. In the Big Bench Audio test, gpt-realtime achieved an 82.8% accuracy, surpassing the 65.6% score of the previous model from December 2024. In the MultiChallenge Audio Benchmark test, the model scored 30.5%, exceeding the previous score of 20.6%.

Along with the new model and voices, there have also been updates to the Realtime API. The API now supports remote MCP servers, visual input, and phone calls via Session Initiation Protocol (SIP). Developers can also save and reuse their prompts.

Despite all these improvements, OpenAI has lowered the price of the Realtime API. gpt-realtime is now 20% cheaper compared to the previous gpt-4o-realtime-preview. The cost is $32 per 1M voice input tokens and $64 per 1M voice output tokens.

You Might Also Like;

Copy URL
URL Copied