Qwen 3

Name: Qwen/Qwen3-1.7B-MLX-4bit
Author: Alibaba

Qwen/Qwen3-1.7B-MLX-4bit

Vendor: Alibaba
Quantization: MLX 4-bit
Parameters: 1.7B
Size: 883.2 MB

$ brew install mirai$ mirai --model Qwen/Qwen3-1.7B-MLX-4bit

Benchmarks

0.32 s

lower is better ↓

0.32 s

uzu

0.39 s

mlx

0.36 s

llamacpp

289 t/s

higher is better ↑

289 t/s

uzu

296 t/s

mlx

194 t/s

llamacpp

1.38 GB

lower is better ↓

1.38 GB

16.02 GB

1.32 GB

0.31.2 llama.cpp0.3.23/1,312 input tokens/512 output tokens

Benchmarked 7 Aug 2026

Integrate with SDK

Choose framework

Run the following command to install Mirai SDK

https://github.com/trymirai/uzu-swift

Apply code

1import Foundation2import Uzu34public func runChat() async throws {5    let engineConfig = EngineConfig.create()6    let engine = try await Engine.create(config: engineConfig)7    8    guard let model = try await engine.model(identifier: "alibaba:qwen3.5:0.8b:mirai:mirai-m:4") else {9        return10    }11    for try await update in try await engine.download(model: model).iterator() {12        print(String(format: "\r\u{001B}[2KDownload progress: %.2f%%", update.progress() * 100), terminator: "")13        fflush(stdout)14    }15    print()16    17    let messages = [18        ChatMessage.system().withText(text: "You are a helpful assistant"),19        ChatMessage.user().withText(text: "Tell me a short, funny story about a robot")20    ]21    let session = try await engine.chat(model: model, config: .create())22    let stream = await session.replyWithStream(input: messages, config: .create())23    var message: ChatMessage? = nil24    for try await update in stream.iterator() {25        switch update {26        case .replies(let replies):27            let reply = replies.last28            message = reply?.message29            print("Generated tokens: \(reply?.stats.tokensCountOutput ?? 0)")30        case .error(let error):31            print("Error: \(error)")32        }33    }34    print("Reasoning: \(message?.reasoning() ?? "empty")")35    print("Text: \(message?.text() ?? "empty")")36}37

1Choose framework

2Run the following command to install Mirai SDK

https://github.com/trymirai/uzu-swift

3Apply code

1	import Foundation
2	import Uzu
3
4	public func runChat() async throws {
5	let engineConfig = EngineConfig.create()
6	let engine = try await Engine.create(config: engineConfig)
7
8	guard let model = try await engine.model(identifier: "alibaba:qwen3.5:0.8b:mirai:mirai-m:4") else {
9	return
10	}
11	for try await update in try await engine.download(model: model).iterator() {
12	print(String(format: "\r\u{001B}[2KDownload progress: %.2f%%", update.progress() * 100), terminator: "")
13	fflush(stdout)
14	}
15	print()
16
17	let messages = [
18	ChatMessage.system().withText(text: "You are a helpful assistant"),
19	ChatMessage.user().withText(text: "Tell me a short, funny story about a robot")
20	]
21	let session = try await engine.chat(model: model, config: .create())
22	let stream = await session.replyWithStream(input: messages, config: .create())
23	var message: ChatMessage? = nil
24	for try await update in stream.iterator() {
25	switch update {
26	case .replies(let replies):
27	let reply = replies.last
28	message = reply?.message
29	print("Generated tokens: \(reply?.stats.tokensCountOutput ?? 0)")
30	case .error(let error):
31	print("Error: \(error)")
32	}
33	}
34	print("Reasoning: \(message?.reasoning() ?? "empty")")
35	print("Text: \(message?.text() ?? "empty")")
36	}
37

Details

A 4-bit quantized version of Qwen3-1.7B optimized for Apple Silicon via the MLX framework. Part of Alibaba's third-generation Qwen language model family, this compact model brings advanced reasoning and conversational capabilities to local Mac-based inference with a minimal memory footprint.

Key Features

Qwen3-1.7B-MLX-4bit inherits the standout capabilities of the Qwen3 series in a lightweight package:

Dual-mode reasoning: Seamlessly switch between a thinking mode (for complex math, coding, and logical reasoning) and a non-thinking mode (for fast, general-purpose dialogue) within a single model. Thinking mode wraps intermediate reasoning in `<think>...</think>` blocks before delivering a final answer.
Multilingual fluency: Supports over 100 languages and dialects, with strong multilingual instruction-following and translation performance.
Agent and tool-calling support: Designed for agentic workflows, including integration with external tools and MCP-compatible servers via frameworks like Qwen-Agent.
Soft switching: Users can toggle thinking behavior mid-conversation using `/think` and `/no_think` commands in prompts, enabling fine-grained control over response style per turn.

Architecture

Type: Causal language model (dense)
Parameters: 1.7B total (1.4B non-embedding)
Layers: 28
Attention: Grouped-Query Attention (16 Q heads, 8 KV heads)
Context length: 32,768 tokens
Quantization: 4-bit (MLX)
Base model: Qwen3-1.7B-Base (pretrained + post-trained)

Use Cases

This model is well-suited for on-device inference on Apple Silicon Macs — ideal for developers and researchers who want local, privacy-preserving access to a capable small language model. It handles creative writing, role-playing, multi-turn dialogue, code generation, and tool-augmented tasks. Compatible with `mlx_lm` (≥ 0.25.2) and `transformers` (≥ 4.52.4).

Explore all local models