OpenAI Group PBC today introduced GPT-Live, a family of artificial intelligence models optimized to process spoken instructions.

The model series will power ChatGPT’s voice mode. Additionally, OpenAI plans to make it available to developers via an application programming interface.

GPT-Live includes two models on launch. GPT-Live-1, the more capable of the two, will be the default option in paid ChatGPT plans while a scaled-down algorithm called GPT-Live-1-mini will power the free tier. The models are rolling out to the iOS, Android and web versions of ChatGPT with support for multiple languages.

ChatGPT’s voice mode previously had to wait until users finished speaking before generating a response. As a result, it wasn’t capable of performing real-time translation. Additionally, the feature could only perform tasks such as searching the web at the end of the user’s monologue even if such tasks could theoretically be completed earlier. That slowed down prompt processing.

According to OpenAI, GPT-Live features a so-called full-duplex architecture that addresses its predecessor’s limitations. During conversations, the model series regularly checks whether it should interrupt the user, search the web or perform some other task. That removes the need for GPT-Live to wait until the user finishes describing a request.

OpenAI developed a set of evaluations to study its new models’ pleasantness. The company says that GPT-Live-1 scored 75.5, which put it well ahead of ChatGPT’s previous voice processing model.

When GPT-Live receives complex prompts, it routes them to GPT-5.5, OpenAI’s most advanced commercially available model. The algorithm set records across several coding benchmarks when it rolled out in April. GPT-Live can use GPT-5.5 to browse the web, explain complex topics and perform a range of other tasks.

OpenAI stated that GPT-5.5 will also be capable of using newer, more advanced reasoning models when they become available. The company reportedly plans to release three such algorithms later this week.

Last month, OpenAI introduced a series of cutting-edge large language models called GPT-5.6. The most advanced LLM in the lineup, Sol, outperforms Claude Mythos 5 in some areas. It debuted alongside two scaled-down algorithms called Tera and Luna that trade off some output quality for lower pricing. 

Axios reported today that OpenAI will make GPT-5.6 broadly available on Thursday after receiving approval to do so from the U.S. government. Currently, GPT-5.6 is only accessible to a limited number of organizations. The White House’s decision to authorize a broader release reportedly came after the Commerce Department’s Center for AI Standards and Innovation carried out an evaluation of the model series.

A few weeks before GPT-5.6’s introduction, OpenAI formed a professional services business called the OpenAI Deployment Company. It focuses on helping organizations adopt the ChatGPT developer’s LLMs. The venture has raised $4 billion in funding from OpenAI and 19 external backers. 

The OpenAI Deployment Company today announced that it’s acquiring a fellow professional services provider called Northslope Inc. Its main specialty is building AI applications atop Palantir Technologies Inc.’s software. According to OpenAI, the company’s revenue grew sevenfold last year.

Image: OpenAI