New real-time audio model targeted for Q1 2026 alongside consumer device ambitions.
Development work is underway on a new artificial intelligence model built specifically for audio generation, according to a report describing an internal initiative led by OpenAI Group PBC. The effort centers on speech output and real-time interaction, with engineering, product, and research functions operating under a single company-led program. No external mandate was cited. The immediate effect of the work, as described, is strategic positioning that places the company closer to direct participation in the consumer electronics market rather than remaining solely a software provider.
Inside the Team Merger Driving OpenAI's Voice Ambitions
Inside the project, multiple internal groups have been consolidated, pulling together engineers, product managers, and research staff previously operating across separate initiatives. The concentration point is audio—speech generation and real-time responsiveness rather than batch or delayed output. Those familiar with the effort described the model as speech-optimized, designed to handle live interaction rather than text-first prompting, with performance characteristics intended for continuous audio exchange that is both fast and persistent.
The initiative is reportedly being led by Kundan Kumar, a former researcher at Character.AI who now heads OpenAI's audio AI efforts.
No legal or regulatory framework was cited in connection with the consolidation. The activity was described as internal and company-directed, without filings, approvals, or external oversight mentioned in the reporting. Instead, the procedural shift focused on how the work was being organized—teams combined, scope narrowed, resources aligned toward a single audio-first objective rather than dispersed experimentation. The model is not framed as a side project; it sits at the center of the initiative.
The target window for release can be the end of March 2026, placing the model squarely within the first quarter of the year. The schedule, as described, ties engineering completion directly to product planning, suggesting the audio system is being built with deployment in mind rather than remaining a research-only artifact.
What the New Model Promises: Natural Speech That Can Interrupt You
The new audio model will reportedly sound more natural, handle interruptions like an actual conversation partner, and even speak while you're talking—something today's models cannot manage. This represents a significant leap from OpenAI's current flagship real-time audio model, GPT-realtime, which uses the transformer architecture but lacks the ability to handle overlapping speech.
The company's current audio capabilities, while impressive, still lag behind its text-based models in speed and accuracy. That shortfall has become a key focus as OpenAI prepares to release its first line of voice-first devices.
The Hardware Pipeline: Smart Glasses, Screenless Speakers, and an AI Pen
Looking ahead, the company plans for an audio-first personal device intended to follow the model's launch, with a timeline of approximately one year from the reporting date. The device concept was framed around voice interaction rather than screens, with exploration underway across several form factors including smart speakers, smart glasses, and a pen-like device operated by voice without a display.
This hardware push gained significant momentum in May 2025 when OpenAI acquired io Products Inc., the startup founded by former Apple design chief Jony Ive, in a deal valued at $6.5 billion. Ive is now taking on "deep creative and design responsibilities across OpenAI," with his team of approximately 55 engineers, scientists, researchers, and product development specialists.
In a joint statement posted on OpenAI's website, Sam Altman said of the partnership: "AI is an incredible technology, but great tools require work at the intersection of technology, design, and understanding people and the world. No one can do this like Jony and his team."
Ive himself has reportedly made reducing device addiction a priority, viewing audio-first design as a chance to address what he sees as the missteps of past consumer gadgets. Ive sees this work as an opportunity to "right the wrongs" of screen-heavy devices.
The first hardware product from OpenAI is rumored to be a contextually aware pen, with manufacturing reportedly being handled by Foxconn in Vietnam rather than China. A separate "to-go" audio device is also in development. These products are being positioned as "third-core" devices meant to complement laptops and smartphones rather than replace them.
No finalized product specifications were disclosed. The descriptions remained at the level of exploration rather than confirmation, listing categories rather than named products. Still, the inclusion of multiple hardware types suggested parallel investigation rather than a single-track bet, with audio as the primary interface and speech as the control layer across each example.
The sequence described placed the audio model first, devices second (software before hardware). The model's real-time speech capabilities are positioned as foundational, enabling hardware designs that rely on continuous voice interaction rather than touch or visual input.
