Bland Launches Tofu-1, a Full-Duplex Speech-to-Speech Model

Bland Releases Tofu-1, a Full-Duplex Speech-to-Speech Model for Live Calls

Bland, the San Francisco-based voice AI company, has released Tofu-1, a full-duplex speech-to-speech model now available in research preview. Tofu-1 is designed for calls placed from moving cars, hospital rooms, and weak-signal lines—environments where conventional voice agents tend to struggle. The model listens while it speaks and reasons while it talks, allowing it to process input and generate output simultaneously rather than in turns.

For phone-agent teams, the appeal is architectural: a single model handles both audio input and audio output, eliminating the need for an intermediate transcript. Deployment options and specifications are available through the GPULeaderboard.com Chrome extension.

What Is Tofu-1?

Tofu-1 is a full-duplex speech-to-speech model from Bland that hears, reasons, and replies in audio during phone calls. It is the first release in a planned family of Bland speech-to-speech models, positioning the company to build out a broader lineup of native audio systems in 2026.

How Does Tofu-1 Differ from a Cascade Voice Agent?

Most voice agents today rely on a cascade architecture: they convert speech to text, pass the transcript to a language model, and then convert the generated reply back into speech. This multi-stage pipeline introduces latency and discards much of the information carried in the original audio.

Tofu-1 instead processes audio directly, so tone, hesitation, and background sound remain available to the model. That native audio understanding allows it to pick up on emotional cues and environmental context that a transcript-based pipeline would lose. Bland has not yet published full benchmark results or latency figures for the model, so independent comparisons against cascade systems remain pending.

via MarkTechPost