back to top
HomeTechOpenAI's New Voice Models Want to Do More Than Talk Back

OpenAI’s New Voice Models Want to Do More Than Talk Back

OpenAI GPT Realtime2 Voice
Mohit Geryani
By Mohit Geryani
0
Last updated:
- Advertisement -

OpenAI is pushing deeper into voice.

The company just launched three new realtime audio models in its API. GPT-Realtime-2 for conversational reasoning, GPT-Realtime-Translate for live multilingual translation, and GPT-Realtime-Whisper for streaming speech transcription.

GPT-Realtime-2 can now handle longer conversations, recover from interruptions more naturally, use tools while someone is still talking, and respond with different reasoning levels depending on the task. OpenAI says the model is designed for things like customer support, scheduling, travel assistance, and other workflows where the AI actually has to keep track of context instead of just replying quickly.

OpenAI is no longer treating voice as a side feature attached to chatbots. It’s starting to position voice as the interface itself. That means live translation during conversations. Real time transcription while meetings are still happening. AI agents that can check your calendar, pull information from apps, or complete actions while the conversation keeps moving.

Voice models are starting to behave more like agents

Three audio models in the OpenAI API

The most interesting part of this launch is not the voices themselves. It’s the fact that OpenAI keeps framing these systems around actions and workflows instead of conversations.

The company highlighted examples like Zillow building voice agents that can search for homes and schedule tours, Deutsche Telekom testing multilingual customer support, and Priceline exploring trip planning that happens conversationally from start to finish.

That points to a shift happening across AI right now. Voice assistants used to exist mainly to answer questions. These new systems are being designed to stay active while tasks are unfolding like checking calendars, updating bookings, pulling information from apps, translating conversations live, or handling interruptions without restarting the interaction.

That’s also why OpenAI focused heavily on realtime reasoning and tool use in this launch. A voice assistant that simply sounds natural is no longer enough. The hard part is making the system useful while the conversation is still moving.

GPT-Realtime-2

Voice changes how people use software

Typing naturally creates pauses. People send a prompt, wait for a response, then move on.

Voice interactions work differently. Conversations keep moving even when requests change halfway through or multiple things happen at once.

That creates a much harder problem for AI systems. The model has to listen continuously, decide when to respond, remember context across longer sessions, and sometimes use tools without interrupting the flow of the conversation itself.

And that’s probably why companies like OpenAI are suddenly investing so heavily in realtime infrastructure.

You May Like: Open-Source TTS Models That Can Clone Voices and Actually Sound Human

GPT-Realtime-Translate may end up being the sleeper feature

The reasoning upgrades will get most of the attention, but the realtime translation model could end up having the bigger commercial impact.

OpenAI says GPT-Realtime-Translate can handle more than 70 input languages and translate into 13 output languages while keeping pace with live conversations. That opens the door for customer support, meetings, events, travel assistance, and sales calls where people no longer need to speak the same language fluently to communicate smoothly.

And unlike older translation systems, OpenAI is clearly pushing for conversations that continue naturally while the translation happens in the background.

The company also highlighted testing from BolnaAI, which said the model handled regional Indian languages like Hindi, Tamil, and Telugu with lower word error rates and fewer fallback failures compared to other systems they tested.

Vimeo is experimenting with the model as well. The company says it’s using GPT-Realtime-Translate for live translation during broadcasts so creators can reach global audiences while streaming in real time. According to Vimeo, one of the biggest improvements was how well the system handled multilingual conversations without breaking flow mid-interaction.

Multilingual voice AI becomes much more useful once it starts handling accents, interruptions, and regional speech patterns reliably in real time.

You May Like: SubQ’s 12M Token Model Could Change How AI Handles Long Context. If It’s Real.

Pricing

All three models are available through OpenAI’s Realtime API.

GPT-Realtime-2 is priced at $32 per million audio input tokens and $64 per million audio output tokens, while GPT-Realtime-Translate costs $0.034 per minute and GPT-Realtime-Whisper costs $0.017 per minute.

Developers can also test the models through OpenAI’s Playground before integrating them into apps and workflows.

Don’t miss any Tech Story

Subscribe To Firethering NewsLetter

You Can Unsubscribe Anytime! Read more in our privacy policy

LEAVE A REPLY

Please enter your comment!
Please enter your name here

YOU MAY ALSO LIKE
Google Built Gemma 4 12B Without Multimodal Encoders

Google Built Gemma 4 12B Without Multimodal Encoders

Mohit Geryani - 0
Every multimodal model you've used has the same basic system. Text goes in one way, images go through a vision encoder first, audio goes through an audio encoder first, and then everything gets handed off to the language model in a form it can work with. The encoders are load-bearing and you don't just remove them.Google actually removed them.Gemma 4 12B takes raw image patches and raw audio waveforms and projects them directly into the same embedding space as text tokens. There is no vision encoder or audio encoder. One decoder handling everything.
MiniMax M3 Shows What Happens When AI Stops Thinking in Turns

MiniMax M3 Shows What Happens When AI Stops Thinking in Turns

Mohit Geryani - 0
Most models quit around submission 30 because they stop finding improvement and exit on their own. That's what happened when MiniMax ran a CUDA kernel optimization task against a field of frontier models. Every model except two called it done within the first 30 submissions. M3's best result came on submission 145. After 24 hours. After multiple plateaus where the numbers stopped moving and a reasonable model would have concluded there was nothing left to find. That's the thing MiniMax released yesterday. An AI model with a 1M token context window, native multimodality, and apparently a problem with knowing when to stop.
Anthropic Files for an IPO. AI Is Entering Its Public Company Era

Anthropic Files for an IPO. AI Is Entering Its Public Company Era.

Mohit Geryani - 0
Anthropic has officially taken its first step toward becoming a public company. In a brief announcement on Monday, the company said it had confidentially submitted a draft S-1 registration statement to the U.S. Securities and Exchange Commission for a proposed initial public offering. The filing doesn't reveal a share price, a fundraising target, or even a timeline. For now, it simply gives Anthropic the option to go public once the SEC review process is complete. Just a few years ago, Anthropic was a small group of former OpenAI researchers trying to build an alternative vision for advanced AI. Today, it sits among the handful of companies shaping the industry's future and that's why this filing matters. It's one of the world's most influential AI labs beginning the transition from a privately funded research company to a business that may eventually answer to public shareholders. For most of the AI boom, the biggest bets were made behind closed doors. Venture firms, sovereign wealth funds, and tech giants supplied the capital while the public watched from the outside. Anthropic's filing suggests that era may be starting to change.