LFM 2.5

Name: LiquidAI/LFM2.5-1.2B-Thinking
Author: LiquidAI

LiquidAI/LFM2.5-1.2B-Thinking

Vendor: LiquidAI
Parameters: 1.2B
Size: 2.2 GB

$ brew install mirai$ mirai --model LiquidAI/LFM2.5-1.2B-Thinking

Benchmarks

0.21 s

lower is better ↓

0.21 s

uzu

0.23 s

mlx

0.24 s

llamacpp

198 t/s

higher is better ↑

198 t/s

uzu

186 t/s

mlx

163 t/s

llamacpp

2.33 GB

lower is better ↓

2.33 GB

18.99 GB

2.23 GB

0.31.2 llama.cpp0.3.23/1,339 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

LFM2.5-1.2B Benchmarks

LFM2.5-1.2B-Thinking is a compact reasoning model from Liquid AI, designed to bring high-quality AI inference to edge devices while running under 1GB of memory. Part of the LFM2.5 family, it builds on the LFM2 hybrid architecture with extended pre-training on 28 trillion tokens and multi-stage reinforcement learning.

Architecture & Specifications

The model features a hybrid design with 1.17B parameters across 16 layers — 10 double-gated LIV convolution blocks paired with 6 grouped-query attention (GQA) blocks. It supports a 32,768-token context window and a vocabulary of 65,536 tokens. Eight languages are supported: English, Arabic, Chinese, French, German, Japanese, Korean, and Spanish.

Liquid AI

Performance & Speed

LFM2.5-1.2B-Thinking rivals much larger models on reasoning benchmarks, achieving strong scores on MATH-500 (87.96), IFEval (88.42), and GSM8K (85.60). On hardware, it delivers 239 tok/s decode on AMD CPU and 82 tok/s on mobile NPU, making it practical for real-time on-device applications.

Recommended Use Cases

Liquid AI recommends this model for agentic tasks, data extraction, and RAG workflows. It also supports structured function calling with Pythonic tool-use syntax. It is less suited for knowledge-intensive tasks or programming.

Deployment Options

The model ships in multiple formats for flexible deployment: native Transformers/vLLM checkpoints, GGUF for llama.cpp and CPU inference, ONNX for cross-platform hardware acceleration, and MLX for Apple Silicon. It also works with LM Studio for local desktop use. Fine-tuning is supported via Unsloth and TRL with LoRA, SFT, DPO, and GRPO recipes.

Explore all local models