25launches

The first launch of Tofu-1.

Tofu-1 page

Tofu-1

Launching today

A speech-to-speech model for noisy, high-stakes phone calls

Hunted by IlyasHunter
Visit bland.ai (opens in a new tab)

Tofu-1 is Bland's first speech-to-speech model for phone agents. It works on the audio itself instead of a transcript, so it can tell the caller from a TV, a crying child or a passenger, hear hesitation, and be interrupted or corrected mid-sentence. It keeps talking while it looks things up, and can generate whole call scenes. Bland says 92% of people in its listening tests couldn't tell it was AI. It is an enterprise research preview with no API or pricing yet; Bland plans a release in January.

Meet the team

Hunted by Ilyas. The makers are not on 25launches yet.

H (Hunter)
  • Tofu-1 screenshot 1
  • Tofu-1 screenshot 2
  • Tofu-1 screenshot 3
  • Tofu-1 screenshot 4

Discussion 1

Ask questions, give feedback. Comments don't count in the ranking.

Sign in to join the discussion.

  • IlyasHunter Pinned

    Phone agents trip on real-world calls: a TV in the room says "yes" and a transcript-based agent hears consent. Tofu-1 listens to the audio itself to work out who is talking to whom, and Bland's "TV Test" write-up explains the problem well. It's only an enterprise research preview and the 92% figure comes from Bland's own tests, but the launch video passed 140k views on X.