Thinking Machines Lab (TML) recently showcased a groundbreaking advancement in artificial intelligence with the introduction of a Full-Duplex Time-aligned micro-turn system, which allows for near-realtime voice and video conversations. This innovation represents a significant shift from traditional turn-based interaction models, enabling continuous, simultaneous communication akin to a phone call rather than a text exchange. TML’s system builds on the foundation laid by MiniCPM-o 4.5 and OpenBMB’s Omni-Flow framework, which focuses on aligning visual, audio, and textual responses within a shared temporal context. This new approach aims to enhance the user experience by allowing AI to perceive and respond in real time, moving away from the outdated walkie-talkie style of communication.

OpenBMB: OpenBMB is an open-source AI research group and software ecosystem focused on building and sharing foundation models and tooling. Here, it is relevant because the news contrasts Thinking Machines Lab’s preview with OpenBMB’s already-shipped full-duplex omni-modal architecture.
Omni-Flow: Omni-Flow is OpenBMB’s framework for synchronizing multimodal inputs and outputs along a shared temporal axis. In this news, it is presented as the technical basis for full-duplex interaction, letting the model perceive and respond at the same time.
MiniCPM-o 4.5: MiniCPM-o 4.5 is an open omni-modal model in the MiniCPM family designed to handle multiple input and output types. The news describes it as an example of a model that already implements continuous, time-aligned interaction across speech, audio, and visual streams.
Qwen3-Omni-30B-A3B: Qwen3-Omni-30B-A3B is a large omni-modal model associated with the Qwen family and designed for multimodal understanding and generation. It is mentioned here as a benchmark that MiniCPM-o 4.5 is said to outperform in omni-modal capability and speech generation quality.
Thinking Machines Lab: Thinking Machines Lab is an AI research and product company founded by former OpenAI CTO Mira Murati. In this news, it is cited for demonstrating an ‘interaction model’ concept built around continuous, real-time AI conversation rather than the older turn-based chat format.

Full-duplex AI: Full-duplex voice systems aim to let users interrupt, talk over, and collaborate with AI more naturally, closer to a phone call than a text exchange.
Interaction models: A growing theme in AI products is moving from chat-style turn taking to continuous, low-latency interaction where the model can listen and respond simultaneously.
Open-model deployment: Recent open omni-modal models are increasingly shipped with code, weights, and edge-friendly deployment options, signaling a push from demos toward practical real-world use.