LFM 2.5

Name: LiquidAI/LFM2.5-1.2B-Instruct-MLX-8bit
Author: LiquidAI

LiquidAI/LFM2.5-1.2B-Instruct-MLX-8bit

Vendor: LiquidAI
Quantization: MLX 8-bit
Parameters: 1.2B
Size: 1.2 GB

$ brew install mirai$ mirai --model LiquidAI/LFM2.5-1.2B-Instruct-MLX-8bit

Benchmarks

0.22 s

lower is better ↓

0.22 s

uzu

0.25 s

mlx

0.24 s

llamacpp

328 t/s

higher is better ↑

328 t/s

uzu

316 t/s

mlx

232 t/s

llamacpp

1.30 GB

lower is better ↓

1.30 GB

18.99 GB

1.32 GB

0.31.2 llama.cpp0.3.23/1,339 input tokens/254 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

Liquid AI

An 8-bit quantized MLX export of Liquid AI's LFM2.5-1.2B-Instruct model, purpose-built for efficient text generation on Apple Silicon hardware. At just 1.2 GB, this model delivers capable instruction-following performance in a highly portable package.

Architecture & Specifications

LFM2.5 is part of Liquid AI's proprietary Liquid Foundation Model family, designed with edge deployment in mind. This variant packs 1.2 billion parameters into an 8-bit quantized format (group size 64), enabling fast local inference via the MLX framework on Mac devices. Despite its compact size, it supports an impressive 128K token context length, making it suitable for tasks that require processing long documents or extended conversations.

Capabilities

The model is multilingual, supporting ten languages including English, Japanese, Korean, French, Spanish, German, Italian, Portuguese, Arabic, and Chinese. As an instruction-tuned model, it handles conversational prompts, question answering, and general-purpose text generation tasks. Liquid AI recommends conservative sampling parameters — low temperature (0.1) and narrow top-p (0.1) — suggesting the model is optimized for precise, deterministic outputs rather than creative generation.

Use Cases

On-device inference on MacBooks, iMacs, and other Apple Silicon machines
Multilingual assistant tasks across ten supported languages
Long-context processing with up to 128K tokens of input
Resource-constrained environments where a sub-2 GB model footprint is essential

Provenance

Built by Liquid AI and exported for MLX using the `mlx-lm` library. The base model is LFM2.5-1.2B-Instruct. Released under the LFM 1.0 License.

Explore all local models