OpenAI has made a new AI voice model called GPT‑Realtime. It is their most advanced voice tool and is made mainly for businesses. At the same time, OpenAI has made its Realtime API open for all developers. The API first came out as a test version in October 2024 and now anyone can use it.
GPT‑Realtime works differently from older AI voice systems. Usually, AI first changes what you say into text, thinks about it, and then changes it back into speech. GPT‑Realtime does not do this. It listens to what you say and replies in voice right away. This makes talking faster and more natural. It is very useful for things like customer help or virtual assistants in companies.
The model has new features. Two new voices have been added: a male one named Cedar and a female one named Marin. The older eight voices are also improved to sound more real. The voices can show emotions and even change tone depending on who is talking. GPT‑Realtime can notice sounds like laughter and answer to them. It can also switch languages in the same conversation.
It is better at understanding numbers and codes, like phone numbers or policy details, in languages such as Chinese, Japanese, French, and Spanish.
GPT‑Realtime can also understand images. Users can upload a picture, and the AI can look at it and talk about it. This is helpful when both pictures and talking are needed. The model can also connect with other systems using something called MCP servers.
In performance tests, GPT‑Realtime scored 82.8 percent on the Big Bench Audio test, which checks how well a voice AI can think and be accurate. The older model from December 2024 scored 65.6 percent, so this is much better.
The price is made for businesses. Developers pay $32 (around ₹2,800) for one million input tokens and $64 (around ₹5,600) for one million output tokens. There is also a cheaper option for saved tokens at $0.40 (about ₹35) per million tokens. This helps companies use the AI without spending too much.
GPT‑Realtime comes at a time when companies like Google, Microsoft, and Amazon are also making AI voice systems better. OpenAI wants conversations to be fast, natural, and expressive. It can use multiple languages and images. The company worked with business partners to test the AI in real situations. They are also working with Anthropic to make the AI safer, showing they care about responsibility and innovation.
GPT‑Realtime is only available through OpenAI’s Realtime API. Developers all over the world can use it now. With faster replies, better voices, understanding images, and higher accuracy, this AI is expected to be a key tool for the next generation of AI business communication.