OpenAI unveils GPT-Realtime-2 with reasoning in live conversation

OpenAI has released three new voice models for its API, letting developers integrate GPT-5-class reasoning directly into audio applications and voice interfaces. The move is another step in the fight for dominance in the AI market.

GPT-Realtime-2: reasoning in real time

GPT-Realtime-2 is the first to bring complex logical inference to live voice conversation. Unlike simple voice assistants, the new model understands the nuance of a conversation's context and can work through multistep tasks without losing the thread. That matters for applications that need consultations, planning, analytics or tech support, where canned template answers simply won't do.

The model processes speech in real time, so users can speak freely without waiting for a pause for processing. Responses arrive at a natural pace, creating the impression of talking to a real person.

Multilingual translation and transcription

OpenAI has released a separate translation model that supports more than 70 input languages. This lets developers build global applications without having to multiply models for each language: one model covers most of the world's population.

OpenAI has also announced a streaming version of Whisper for transcription. It processes audio in real time and outputs text as the sound arrives. That is critical for applications such as video calls, live translators and speech assistants, where latency directly affects UX.

The three key components:

• GPT-Realtime-2 for speech reasoning and dynamic dialogue
• A translation model supporting 70+ input languages
• Streaming Whisper for low-latency audio transcription

Pricing strategy: grabbing the market

OpenAI has set aggressive prices for the new models, making them accessible to small developer teams and startups. The company is clearly aiming to quickly capture a share of the voice AI application market.

This approach contrasts with the positioning of its text models, where OpenAI holds a premium price position. Investing in the affordability of voice models suggests OpenAI sees voice as the next frontier of AI interaction. Whoever wins over developers in this space first will have a strong competitive advantage.

What this means

Voice AI interfaces are moving from the experimental phase to a practical part of the developer's stack. More affordable pricing lowers the barrier to entry: a startup can now build speech AI into its app without a large investment. That will speed up the arrival of new voice applications on the market.