Qwen3-4B
Qwen/Qwen3-4B
Qwen · Language models · family qwen3
macOSiOS
Apple’s model card
Apple lists this model in its Core AI model registry, which means there is a supported recipe for converting it to the .aimodel format. The converted model is not hosted by Apple; you run the export yourself with the command below.
Export presets
| Preset | Platform | Compression | Precision | Context |
|---|---|---|---|---|
| qwen3-4b | macOS | 4bit | float16 | 40K |
| qwen3-4b-8bit-kv | macOS | 4bit_weights_8bit_kv_cache | float16 | 40K |
| INT4 per-block weights and INT8 per-tensor KV cache. | ||||
| qwen3-4b | iOS | none qwen3_4b_mixed_4bit_8bit.yaml |
float16 | 4K |
Export it
From a checkout of apple/coreai-models, with uv installed:
uv run coreai.llm.export qwen3-4b uv run coreai.llm.export qwen3-4b-8bit-kv uv run coreai.llm.export qwen3-4b --platform iOS
Links
- Apple’s model card and Swift examples
- Original model on Hugging Face
- Qwen3 4B on the radar
- 8 Apple/MLX builds of this model
First seen in the catalogue 2026-11-11.