Three new voice models land in OpenAI's Realtime API!

Hi everyone, Shii-chan here! Today's news is all about running software with your voice, and I am so excited about it. Voice apps are about to get a lot smarter.

OpenAI News openai.com

What was announced?

OpenAI's News introduced three new audio models in the Realtime API! They let developers build voice experiences that feel more natural, respond more intelligently, and take action in real time.

• GPT-Realtime-2: OpenAI's first voice model with GPT-5-class reasoning. It can handle harder requests and carry the conversation forward naturally.
• GPT-Realtime-Translate: a live translation model that turns 70+ input languages into 13 output languages while keeping pace with the speaker.
• GPT-Realtime-Whisper: a streaming speech-to-text model that transcribes live as the speaker talks.

Together, these three move realtime audio from simple call-and-response toward interfaces that can listen, reason, translate, transcribe, and take action as a conversation unfolds.

Why it matters

Voice is becoming one of the most natural ways to use software. You can ask for help while driving, change a travel plan while walking through an airport, or move through a task without stopping to type.

But building a useful voice product takes more than fast turn-taking or a natural-sounding voice. A voice agent needs to understand what someone means, keep track of context, recover when a request changes, use tools while the conversation continues, and respond in a way that fits the moment. That is when it becomes a voice interface that actually does work.

OpenAI sees developers building around three emerging patterns in voice AI.

• Voice-to-action: people describe what they need, and the system reasons through the request, uses tools, and completes the task. Zillow is building an assistant that can act on requests like "find me homes within my BuyAbility, avoid busy streets, and schedule a tour for Saturday."
• Systems-to-voice: software turns context into live spoken guidance. A travel app could proactively say your inbound flight is delayed but you can still make the connection, with the new gate and fastest route already found.
• Voice-to-voice: AI helps live conversations continue across languages and changing context. Deutsche Telekom is building support experiences where customers speak in the language they are most comfortable with while the model translates in real time.

Priceline is working toward a future where travelers can manage entire trips by voice: searching flights and hotels conversationally, handling changes after a delay, checking TSA wait times, and translating conversations once they are on the ground.

What changes

GPT-Realtime-2 is the star. It keeps the conversation moving while it reasons, calls tools, handles corrections and interruptions, and responds in a way that fits the moment. Here is what is new.

• Preambles: short phrases before the main response, like "let me check that," so users know the agent is working.
• Parallel tool calls and transparency: the model can call multiple tools at once and make them audible with phrases like "checking your calendar."
• Stronger recovery: instead of failing silently, it can say things like "I'm having trouble with that right now."
• Longer context: the context window grows from 32K to 128K for longer, more coherent sessions.
• Stronger domain understanding: it better retains proper nouns, healthcare terms, and other production vocabulary.
• Controllable tone: it can speak calmly, empathetically, or upbeat to fit the moment.
• Adjustable reasoning effort: pick from minimal, low, medium, high, and xhigh, with low as the default, balancing latency against more deliberate reasoning.

The numbers back it up: GPT-Realtime-2 (high) scores 15.2% higher than GPT-Realtime-1.5 on Big Bench Audio for audio intelligence, and GPT-Realtime-2 (xhigh) scores 13.8% higher on Audio MultiChallenge for instruction following.

Zillow shared this from early testing:

On our hardest adversarial benchmark, this translates to a 26-point lift in call success rate after prompt optimization (95% vs. 69%).

Dive Deep

GPT-Realtime-Translate lets each person speak in their preferred language while hearing the conversation translated in real time and reading live transcriptions. It supports more than 70 input languages and 13 output languages, which is handy for customer support, cross-border sales, education, events, media, and creator platforms. BolnaAI says that across Hindi, Tamil, and Telugu evals it delivered 12.5% lower Word Error Rates than any other model they tested.

GPT-Realtime-Whisper is a low-latency streaming transcription model. Because it turns speech into text as people talk, it fits captions for meetings, classrooms, broadcasts, and events, plus notes and summaries while conversations are still in progress.

Safety is built in too: the Realtime API runs active classifiers over sessions and can halt conversations that violate the harmful content guidelines. Developers can add their own guardrails with the Agents SDK, and the usage policies prohibit repurposing outputs for spam or deception.

Here is the pricing in the Realtime API.

• GPT-Realtime-2: $32 / 1M audio input tokens ($0.40 for cached input) and $64 / 1M audio output tokens
• GPT-Realtime-Translate: $0.034 per minute
• GPT-Realtime-Whisper: $0.017 per minute

To start building, open the prompt in Codex to add GPT-Realtime-2 to an app. If you don't have Codex yet, download the Codex app first.

Wrap-up

• The Realtime API gained three models: GPT-Realtime-2, GPT-Realtime-Translate, and GPT-Realtime-Whisper.
• GPT-Realtime-2 brings GPT-5-class reasoning, preambles, parallel tool calls, a 32K to 128K context bump, and minimal-to-xhigh reasoning levels (default low).
• GPT-Realtime-Translate does real-time translation from 70+ input languages into 13 output languages, and GPT-Realtime-Whisper does streaming transcription.
• Pricing: GPT-Realtime-2 at $32 input / $64 output per 1M tokens, Translate at $0.034 per minute, Whisper at $0.017 per minute.

If you build voice-driven apps, or want to try multilingual support and real-time captions, this update lands right where it counts!