Block's AI lab releases Dessa, a full-duplex speech-to-speech model that runs on a MacBook Pro M4 Max
Block's AI lab released Dessa, a full-duplex speech-to-speech model trained only on human conversation audio. It runs locally on a MacBook Pro M4 Max, according to MarkTechPost.
Block's AI lab has released Dessa, a full-duplex speech-to-speech model trained only on human conversation audio, according to a MarkTechPost report published October 29, 2026. The model runs locally on a MacBook Pro M4 Max.
What is confirmed
- Model: Dessa, from Block's AI lab.
- Type: Full-duplex speech-to-speech.
- Training data: Human conversation audio only.
- Local hardware: Runs on a MacBook Pro M4 Max.
MarkTechPost reports that Block has published no model size, benchmarks or weights.
What this means
Dessa pairs two design choices: full-duplex operation and training on speech alone. Running on a MacBook Pro M4 Max means the model is described as working on a single laptop-class machine. Without a published parameter count, benchmark results or weights, its quality and efficiency cannot be compared with other speech models, and outside developers cannot yet run or test it themselves.
Related Articles
Liquid AI releases LFM3-VL-2B, a 1.92B on-device vision-language model with video input and thinking
Liquid AI has released LFM3-VL-2B, a 1.92B-parameter vision-language model that accepts text, images and video. It is the first LFM vision model with video input, adds step-by-step thinking, and ships with a 128K-token context window under the LFM Open License v1.0.
AssemblyAI releases AssemblyAI-TTS-1 text-to-speech model at $15 per 1M characters
AssemblyAI has released AssemblyAI-TTS-1, its first text-to-speech model for voice agents. It costs $15 per 1M characters, offers unlimited concurrency without a contract, and supports 6 languages at launch.
Diffusion Intelligence releases Clover-35B-A3B-Preview, a 35B-parameter diffusion language model for agents
Diffusion Intelligence (扩散智能) has released Clover-35B-A3B-Preview, a mixture-of-experts diffusion language model with 35B total and about 3B active parameters per token. The team reports 73.6 on SWE-bench Verified and a 1.52-second median task time on the WindTunnel web-agent benchmark.
NVIDIA Nemotron 3.5 Bolt E4B: 8B-parameter Gemma 4 post-train with 64K context for local assistants
NVIDIA released Nemotron-3.5-Bolt-E4B-4GB on 2026-10-23, a small language model post-trained from Gemma-4-E4B-it with 8B total (4B effective) parameters and 64K-token input and output. It targets local voice and device assistants. Unsloth now lists a GGUF build of the model on Hugging Face.
Comments
Loading...