Qwen3-TTS 1.7B Apple Silicon Benchmarks
The default voice model. 4-bit, about 2.3 GB on disk and around 11 GB of memory while it runs.
Who makes it
Qwen3-TTS is Alibaba's speech line. The 12 Hz in every repo name is its audio token rate. GroundSlate installs four checkpoints for two jobs: a Base checkpoint speaks your scripts, and a CustomVoice or VoiceDesign checkpoint makes a voice once and is not used after that.
- Maker
- Alibaba
- Line
- Qwen3-TTS
- Size
- 2.3 GB
- Weights
- mlx-community/Qwen3-TTS-12Hz-1.7B-Base-4bit
- License
- Apache License 2.0
Made with Qwen3-TTS 1.7B
Made on the reference Mac from the brief shown with each one. Every sample has its own page, and you can run each prompt on any model that accepts it.
Speed
On each chip
How long Qwen3-TTS 1.7B takes on every chip, timed by the benchmark in the app on customers' own Macs. Every test on every chip
Spoken line
median timeSpeaks one fixed sentence with the local voice model.
- M632 GBFastest3.21s3.0× faster than real time
- M3 Max64 GB3.43s3.0× faster than real time
MacBook Pro, Apple M3 Max, 64 GB unified memory. 2026-08-28.