Bland Launches Tofu 1 Full Duplex Speech Model for Phone Calls

Executive Briefing
“Voice AI company Bland has introduced Tofu 1, a full duplex speech to speech model designed...”
Key Takeaways
- 1.The Tofu 1 model processes audio directly to retain vocal tone, hesitation and background sounds without converting text first.
- 2.The system supports full duplex operations allowing callers to interrupt mid sentence while the model maintains conversation and reasons through requests.
- 3.Bland released the model as a closed research preview targeting enterprises in banking, healthcare, insurance and government sectors.
Voice AI company Bland has introduced Tofu 1, a full duplex speech to speech model designed for telephone calls in a research preview.
Key Developments and Technical Context
The Tofu 1 model processes audio directly to retain vocal tone, hesitation and background sounds without converting text first.
The system supports full duplex operations allowing callers to interrupt mid sentence while the model maintains conversation and reasons through requests.
Bland released the model as a closed research preview targeting enterprises in banking, healthcare, insurance and government sectors.
Strategic Analysis and Industry Implications
Removing intermediate text conversion helps the model capture nonverbal cues like emotion and background noise in realistic environments more effectively.
The ability to process tasks while conversing reduces conversational latency, delivering a more natural dialogue experience than traditional pipeline systems.
Outlook and Key Factors to Watch
Bland has not disclosed the internal architecture, parameter size, specific latency metrics or commercial pricing details of the model.
The actual performance of the system under weak cellular network conditions remains unverified independently outside of self reported developer metrics.







