AIToday
Audio & SpeechAI Business & IndustryOpenAI BlogPublished: May 8, 2026, 04:00 JST

OpenAI launches three new realtime voice models

Send on LINE

3 Key Points

  1. What happened

    OpenAI introduced GPT-Realtime-2, GPT-Realtime-Translate, and GPT-Realtime-Whisper in its API. GPT-Realtime-2 is its first voice model with GPT-5-class reasoning and boosts the context window from 32K to 128K tokens.

  2. Why it matters

    Voice is becoming a key way people use software, but past voice agents struggled with understanding context, handling corrections, and taking action. These models are designed to listen, reason, translate, transcribe, and act in real time, with GPT-Realtime-2 (high) scoring 15.2% higher on Big Bench Audio than GPT-Realtime-1.5.

  3. What to watch

    Adoption by companies like Zillow, Deutsche Telekom, and Priceline will show whether these models handle complex real-world tasks. GPT-Realtime-2 pricing is $32 / 1M audio input tokens and $64 / 1M audio output tokens.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Context & Analysis

OpenAI's new API models address a core limitation of earlier voice AI: they were often rigid, failing to understand changing context or perform complex tasks. The company's focus on 'voice-to-action', 'systems-to-voice', and 'voice-to-voice' patterns suggests a push toward making voice a primary interface for everyday software use, not just a novelty.

The launch builds on previous realtime voice models, but the key advance is combining conversational ability with reasoning and action, exemplified by GPT-Realtime-2's GPT-5-class intelligence. The expansion of context from 32K to 128K tokens is also notable, as it enables longer and more complex interactions. The inclusion of prominent partners like Zillow and Deutsche Telekom indicates OpenAI prioritizes moving these models from demos to practical business workflows.

The success of these models will likely hinge on whether they can reliably handle the complexities of real-world speech—accents, interruptions, and ambiguous requests—while maintaining acceptable latency and cost. The clear pricing structure and a demo suggest OpenAI wants rapid developer feedback to iterate, but the market's response will determine if this marks a turning point for voice AI or just another incremental step.

FAQ
What is the difference between the three new models?
GPT-Realtime-2 focuses on real-time conversational reasoning and tool use, GPT-Realtime-Translate performs live speech translation, and GPT-Realtime-Whisper offers low-latency streaming transcription.
How much do the new models cost?
GPT-Realtime-2 is priced at $32 / 1M audio input tokens and $64 / 1M audio output tokens. GPT-Realtime-Translate is $0.034 per minute, and GPT-Realtime-Whisper is $0.017 per minute.
Which companies are already testing these models?
Zillow is building an assistant for real estate tasks, Deutsche Telekom is testing multilingual voice support, Priceline is working on voice-managed trips, and Vimeo is demonstrating live video translation.
Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleCommerce.com reports 5% revenue growth and returns to GAAP profitability in Q1 2026 as AI-driven commerce initiatives expand