---
title: Alibaba Apple Silicon Benchmarks
summary: Vision-language and speech weights from Alibaba's Qwen team. GroundSlate runs the MLX conversions from mlx-community: Qwen3-VL describes footage, Qwen3-TTS speaks.
---

# Alibaba Apple Silicon Benchmarks

Vision-language and speech weights from Alibaba's Qwen team. GroundSlate runs the MLX conversions from mlx-community: Qwen3-VL describes footage, Qwen3-TTS speaks.

- Site: https://qwenlm.github.io

## Qwen3-VL

Qwen3-VL is Alibaba's vision-language line. GroundSlate runs the 4-bit MLX conversions from mlx-community.

- Page: /resources/groundslate/benchmarks/makers/alibaba/qwen3-vl

- Qwen3-VL 4B (`mlx-community/Qwen3-VL-4B-Instruct-4bit`): still 3.12s, video chunk 14.14s. /resources/groundslate/benchmarks/vision/qwen3-vl-4b
- Qwen3-VL 8B (`mlx-community/Qwen3-VL-8B-Instruct-4bit`): still 5.89s, video chunk 24.72s. /resources/groundslate/benchmarks/vision/qwen3-vl-8b

## Qwen3-TTS

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.

- Page: /resources/groundslate/benchmarks/makers/alibaba/qwen3-tts

- Qwen3-TTS 1.7B (`mlx-community/Qwen3-TTS-12Hz-1.7B-Base-4bit`): one line 8.57s, six lines 14.86s. /resources/groundslate/benchmarks/voice/qwen3-tts-1-7b