Qwen 3

Name: Qwen/Qwen3-1.7B-MLX-8bit
Author: Alibaba

Qwen/Qwen3-1.7B-MLX-8bit

Vendor: Alibaba
Quantization: MLX 8-bit
Parameters: 1.7B
Size: 1.7 GB

$ brew install mirai$ mirai --model Qwen/Qwen3-1.7B-MLX-8bit

Benchmarks

0.32 s

lower is better ↓

0.32 s

uzu

0.39 s

mlx

0.36 s

llamacpp

192 t/s

higher is better ↑

192 t/s

uzu

195 t/s

mlx

127 t/s

llamacpp

2.18 GB

lower is better ↓

2.18 GB

16.02 GB

2.50 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

Qwen3-1.7B-MLX-8bit is an 8-bit quantized version of Qwen's third-generation 1.7B-parameter language model, optimized for Apple Silicon via the MLX framework. It delivers the full capabilities of Qwen3 in a compact, efficient package suited for on-device inference on Mac hardware.

Architecture & Specifications

Built on a causal language model architecture, Qwen3-1.7B features 28 layers with grouped-query attention (16 Q heads, 8 KV heads) and supports a context length of 32,768 tokens. The base model contains 1.7B total parameters (1.4B non-embedding), making it one of the most lightweight entries in the Qwen3 lineup. The 8-bit MLX quantization further reduces memory footprint while preserving quality.

Thinking and Non-Thinking Modes

A standout feature of Qwen3 is its ability to seamlessly switch between thinking mode — where the model reasons step-by-step through complex math, logic, and coding problems — and non-thinking mode, which provides fast, direct responses for general conversation. This can be toggled via `enable_thinking` in the chat template, or dynamically controlled mid-conversation using `/think` and `/no_think` tags in user messages.

Key Capabilities

Reasoning: Enhanced performance on mathematics, code generation, and commonsense reasoning tasks compared to prior Qwen generations.
Agent & Tool Use: Strong integration with external tools and MCP-based workflows, supported natively through Qwen-Agent.
Multilingual: Supports over 100 languages and dialects, with robust instruction-following and translation abilities.
Conversational Quality: Improved alignment for creative writing, role-playing, multi-turn dialogue, and instruction following.

Ideal Use Cases

This model is well-suited for developers building lightweight, on-device AI applications on macOS — particularly those needing a balance of reasoning depth and fast inference. It requires `mlx_lm` ≥ 0.25.2 and `transformers` ≥ 4.52.4. Released under the Apache 2.0 license by the Qwen Team at Alibaba.

Explore all local models