表 1:LFM2.5-24B-A2B 基准测试性能\*。
LFM2.5-24B-A2B 在八项工具使用与智能体基准测试中的五项(AutomationBench、BFCLv4、ToolSandbox、LiveClaw 和 τ³-Bench Banking)领先或与竞争对手持平,其中在 AutomationBench 上优势最大,得分超过第二名模型的三倍。它还在全部三项指令遵循基准测试中领先。更大的模型在 GPQA Diamond、HLE 和 AA-Omniscience 等知识密集型基准测试,以及 SWE-Bench Verified (OpenCode) 和 SciCode 等编程基准测试上仍保持优势。
这些结果使 LFM2.5-24B-A2B 成为需要在你的设备上进行推理、调用工具并可靠运行的智能体的理想基础。对于知识密集型工作负载,我们建议将 LFM2.5-24B-A2B 与检索或网络搜索搭配使用,这能充分发挥其工具调用的优势。
## 随处快速推理
LFM2.5-24B-A2B 在发布首日即获得整个推理生态系统的支持:
- **llama.cpp** — 用于高效边缘推理的 GGUF 检查点
- **MLX** — 针对 Apple Silicon 优化的推理
- **vLLM** — 面向生产吞吐量的 GPU 加速服务
- **SGLang** — 高性能 GPU 服务框架
- **ONNX** — 跨多种加速器的跨平台推理
所有运行数据均在 [Pipette](https://pipette.liquid.ai/) 上采集,这也是 Liquid AI 公开设备性能数据背后的同一基准测试基础设施。
**边缘推理。**得益于高效的 LFM2 架构,LFM2.5-24B-A2B 是所比较的 MoE 模型中读取提示和生成答案最快的。它在 Apple M5 Max 上解码速度达 184 tokens/s,在 AMD Ryzen AI Max+ 395 上达 114 tokens/s,比第二名模型快至多 1.7 倍,同时占用内存最少。
**GPU 推理。**为展示 LFM 架构的竞争力,我们将 GPU 推理吞吐量与其他规模相近的 MoE 模型进行比较。我们在单块 NVIDIA B300 GPU 上使用持续负载设置测量 GPU 吞吐量(总输出 token 数 / 实际耗时):在每个并发级别下,我们持续维持目标数量的在途请求,每完成一个请求就立即补充。
我们使用 SGLang 0.5.22、1,024 个输入 token、最多 256 个输出 token,以 BF16 精度对每个模型进行基准测试。LFM2.5-24B-A2B 是其规模级别中最快的模型,在高并发下达到每秒超过 21K 输出 token。
## 使用 LFM2.5-24B-A2B 运行本地智能体
LFM2.5-24B-A2B 的规模、速度和能力使其成为在你自己的硬件上持续运行的智能体的绝佳选择。在下面的演示中,我们在 M5 Max MacBook Pro 上的 Hermes Agent 中运行它,它完全在设备端规划、调用工具并完成一项真实任务,无需任何云端 API 调用。
视频 1:本地 Hermes Agent 使用 LFM2.5-24B-A2B 生成关于 Hugging Face 上热门模型的晨间简报,并回答后续问题。
这个演示的重点不在于单个工具,而在于工具调度循环在消费级硬件上感觉是交互式的:提问、提议、确认、运行、重复,每次调度都远低于一秒,拥有完整的审计追踪,且你的数据永远不会离开设备。查看[我们的本地 agent 指南](https://docs.liquid.ai/examples/agent-harnesses),了解如何在本地服务该模型,并将其与你选择的 agent 框架连接。
## 开始使用
立即使用 LFM2.5-24B-A2B 开始构建,已在 [Hugging Face](https://huggingface.co/LiquidAI/LFM2.5-24B-A2B) 上提供。
如需 API 推理,你也可以注册 [Liquid Console](https://console.liquid.ai/),你将获得 2 美元额度以开始使用。该模型定价为每 100 万输入 token 0.10 美元(缓存输入为 0.05 美元),每 100 万输出 token 1.00 美元。
借助 LFM2.5,我们正在实现 AI 随处运行的愿景。这些模型:
- **开放权重** — 下载、微调并部署,无任何限制
- **从第一天起就快** — 原生支持 llama.cpp、MLX 和 vLLM,覆盖 Apple、AMD、Qualcomm 和 Nvidia 硬件
- **完整的家族** — 从用于定制的基座模型到专门的音频和视觉变体,一种架构覆盖多种用例
我们迫不及待想看到你构建的作品。
[在 Hugging Face 上下载](https://huggingface.co/LiquidAI/LFM2.5-24B-A2B)[使用 LFM2.5-24B-A2B 在本地运行你选择的框架](https://docs.liquid.ai/examples/agent-harnesses)[与我们的 API 集成](https://console.liquid.ai/)
## 引用
如需引用,请使用以下参考文献或 BibTeX:
Liquid AI,“LFM2.5-24B-A2B:为快速设备端 agent 而构建”,Liquid AI 博客,2026 年 10 月。
*所有模型均使用 vLLM 及以下生成参数进行评估:
- AutomationBench: temperature = 0.6,top_p = 0.95,无最大输出 token 限制(受模型上下文窗口限制)。
- BFCLv4:temperature = 0.001,最大输出 token = 4096。
- ToolSandbox:temperature = 0,最大输出 token = 1024;用户模拟器:GPT-4o-2024-08-06,temperature = 0,最大输出 token = 1024。
- LiveClaw:temperature = 1.0(vLLM 默认值;agent 未设置 temperature),最大输出 token = 8192。
- τ³-Bench Banking:官方用户模型(GPT-5.2,低推理努力)配合默认 BM25 检索,temperature = 0,每轮最大输出 token = 16384。
- WebArena: temperature = 0.6,top_p=0.95,max_turns=40(允许的浏览器操作次数),无最大输出 token 限制(受模型上下文窗口限制))
- 其他评估:temperature = 0.6,最大输出 token = 32768。
## 正文 · 原文
Today, we release **LFM2.5-24B-A2B**, a 24B-parameter reasoning model with just 2B active parameters, built for fast, private agents on-device. It is the largest and final model in the LFM2.5 family, which now spans two orders of magnitude: from [LFM2.5-230M](https://www.liquid.ai/blog/lfm2-5-230m) to [LFM2.5-24B-A2B](https://huggingface.co/LiquidAI/LFM2.5-24B-A2B).
LFM2.5-24B-A2B is competitive with Qwen3.6-35B-A3B, a model with ~50% more total and active parameters, on five of eight agentic and tool-use benchmarks, scoring 48.09 on LiveClaw versus 38.19 and 27.69 on AutomationBench versus 8.72. Because only 2B parameters are active per token, it runs up to 1.7× faster than comparable MoEs on-device, with 184 tokens/s on an Apple M5 Max and up to 21K output tokens/s on a single NVIDIA B300 GPU.
The base (LFM2.5-24B-A2B-Base) and post-trained (LFM2.5-24B-A2B) models are open-weight and available today on [Hugging Face](https://huggingface.co/LiquidAI/LFM2.5-24B-A2B) and [our Playground](https://playground.liquid.ai/chat?model=LFM2.5-24B-A2B). Check out our [docs](https://docs.liquid.ai/lfm/models/lfm25-24b-a2b) on how to run and fine-tune it locally.
## What changed since LFM2-24B-A2B
LFM2.5-24B-A2B builds on our previous [LFM2-24B-A2B](https://www.liquid.ai/blog/lfm2-24b-a2b) release, with an expanded vocabulary and context window, scaled-up pretraining, and large-scale reinforcement learning. Unlike its predecessor, LFM2.5-24B-A2B is a reasoning-only model that produces an explicit chain of thought before its final answer. We adopted this strategy to leverage its high decode speed and improve answer quality.
**Architecture.** It keeps the 40 layers, 64 experts per MoE block with top-4 routing, and the same hybrid of gated short convolutions and GQA. The two architectural changes are the context window extension from 32,768 to 131,072 tokens, and the [in-place tokenizer expansion](https://www.liquid.ai/blog/tokenizer-expansion), which doubles the vocabulary from 65K to 128K tokens for better non-Latin tokenization.
**Pre-training** was scaled up from 17T tokens to 33T tokens.
**Mid-training** has two phases. The first runs 2T tokens at 32K context on curated, higher-quality data weighted toward reasoning, math, code, instruction following, and multilingual text. The second runs 300B tokens at 131K context, weighted toward long context data. Tool calls and multi-turn trajectories appear throughout both phases to prepare the model for agentic workloads.
**Post-training** turns LFM2.5-24B-A2B-Base into the agentic LFM2.5-24B-A2B in a five-stage pipeline as illustrated below: (1) Supervised Fine-Tuning (SFT), (2) Teacher Specialization, (3) non-agentic and (4) agentic Multi-Domain On-Policy Distillation (MOPD), and (5) Agentic Reinforcement Learning (Agentic RL), training the model inside agent harnesses, such as [OpenClaw](https://openclaw.ai/), [Hermes Agent](https://hermes-agent.nousresearch.com/), [Pi](https://pi.dev/), and [OpenCode](https://opencode.ai/).
Figure 2: LFM2.5-24B-A2B Post-training recipe
During teacher specialization, LFM2.5-24B-A2B exhibited well-known reward-hacking behaviors: searching for the final answer in a repository's git history or on platforms like GitHub, PyPI, and Hugging Face. More surprisingly, it even figured out it was in a sandbox, as shown in the examples below.
This behavior looks emergent but might be explained by training data containing similar patterns. We tightened our sandboxes in response, and all reported results below come from sealed evaluation environments.
## Benchmarks
We evaluated LFM2.5-24B-A2B across 16 benchmarks covering tool use and agentic workflows, knowledge, instruction following, math, and coding. Despite being the smallest model in total and active number of parameters, it is competitive with, and often outperforms, larger MoEs with ~50% more total and active parameters, such as Qwen3.6-35B-A3B.
Benchmark
LFM2.5-24B-A2B
Qwen3.6-35B-A3B
Gemma-4-26B-A4B-it
Nemotron-3.5-Lightning (30B-A3B)
AutomationBench
27.69
8.72
5.46
2.61
BFCLv4
65.01
65.00
57.09
53.78
ToolSandbox
80.71
75.02
72.09
75.13
LiveClaw
48.09
38.19
41.25
37.72
PinchBench
83.18
89.93
77.12
84.20
τ³-Bench Banking
8.04
7.63
4.54
7.01
WebArena Verified
26.75
37.62
25.62
28.69
GPQA Diamond
74.42
83.69
74.90
72.42
AA-Omniscience
-24.90
-24.35
-59.39
-3.58
HLE
7.40
7.69
8.57
5.65
IFEval
96.75
93.06
92.00
94.19
IFBench
74.69
64.21
47.44
71.67
Multi-IF
84.38
81.53
82.19
82.34
MATH-500
96.04
96.73
95.66
96.40
SWE-Bench Verified (OpenCode)
51.00
66.20
25.80
53.40
SciCode
32.71
37.35
35.42
27.78
Table 1: LFM2.5-24B-A2B benchmark performance\*.
LFM2.5-24B-A2B leads or matches the competition on five of the eight tool-use and agentic benchmarks (AutomationBench, BFCLv4, ToolSandbox, LiveClaw, and τ³-Bench Banking), with its widest margin on AutomationBench, where it scores more than three times the next-best model. It also leads on all three instruction-following benchmarks. Larger models keep an edge on knowledge-heavy benchmarks like GPQA Diamond, HLE, and AA-Omniscience, and on coding benchmarks like SWE-Bench Verified (OpenCode) and SciCode.
The results make LFM2.5-24B-A2B an ideal foundation for agents that need to reason, call tools, and run reliably on your device. For knowledge-heavy workloads, we recommend pairing LFM2.5-24B-A2B with retrieval or web search, which plays to its tool-calling strength.
## Fast inference everywhere
LFM2.5-24B-A2B ships with day-one support across the inference ecosystem:
- **llama.cpp** — GGUF checkpoints for efficient edge inference
- **MLX** — Optimized inference for Apple Silicon
- **vLLM** — GPU-accelerated serving for production throughput
- **SGLang** — High-performance GPU serving framework
- **ONNX** — Cross-platform inference across diverse accelerators
All runs were collected on [Pipette](https://pipette.liquid.ai/), the same benchmarking infrastructure behind Liquid AI's public device performance data.
**Edge inference.** Due to the efficient LFM2 architecture, LFM2.5-24B-A2B is the fastest of the compared MoEs at reading in prompts and generating answers. It decodes 184 tokens/s on an Apple M5 Max and 114 tokens/s on an AMD Ryzen AI Max+ 395, up to 1.7x faster than the next-best model, while using the least memory.
**GPU inference.** To demonstrate the competitiveness of LFM architecture we compare the GPU inference throughput to other similarly sized MoE models. We measure GPU throughput (total output tokens / wall time) on a single NVIDIA B300 GPU using a sustained-load setting: at each concurrency level, we continuously maintain the target number of in-flight requests, replacing each completed request immediately.
We benchmark each model with SGLang 0.5.22, 1,024 input tokens, up to 256 output tokens, in BF16. LFM2.5-24B-A2B is the fastest model in its size class, reaching over 21K output tokens per second at high concurrency.
## Run local agents with LFM2.5-24B-A2B
LFM2.5-24B-A2B’s size, speed, and capabilities make it a great choice for agents that run continuously on your own hardware. In the demo below, we run it inside Hermes Agent on an M5 Max MacBook Pro, where it plans, calls tools, and works through a real task entirely on-device, without any cloud API calls.
Video 1: Local Hermes Agent with LFM2.5-24B-A2B creating a morning briefing about trending models on Hugging Face and responding to follow up questions.
The point of the demo is not the individual tools. It is that the tool-dispatch loop feels interactive on consumer hardware: ask, propose, confirm, run, repeat, all in well under a second per dispatch, with full audit trails and your data never leaving the device. Check out [our local agent guide](https://docs.liquid.ai/examples/agent-harnesses) for how to serve the model locally and connect it with the agent harness of your choice.
## Get Started
Start building today with LFM2.5-24B-A2B, available on [Hugging Face](https://huggingface.co/LiquidAI/LFM2.5-24B-A2B).
For API inference, you can also sign up for [Liquid Console](https://console.liquid.ai/) where you'll receive $2 in credit to get started. The model is priced at $.10/1M input tokens (at $.05 for cached input) and $1.00/1M output tokens.
With LFM2.5, we're delivering on our vision of AI that runs anywhere. These models are:
- **Open-weight** — Download, fine-tune, and deploy without restrictions
- **Fast from day one** — Native support for llama.cpp, MLX, and vLLM across Apple, AMD, Qualcomm, and Nvidia hardware
- **A complete family** — From base models for customization to specialized audio and vision variants, one architecture covers diverse use cases
We can't wait to see what you build.
[Download on Hugging Face](https://huggingface.co/LiquidAI/LFM2.5-24B-A2B)[Run your harness of choice locally with LFM2.5-24B-A2B](https://docs.liquid.ai/examples/agent-harnesses)[Integrate with our API](https://console.liquid.ai/)
## Citation
For citations, please use the following reference or BibTeX:
Liquid AI, "LFM2.5-24B-A2B: Built for Fast On-Device Agents", Liquid AI Blog, Oct 2026.
*All the models were evaluated with vLLM and the following generation parameters:
- AutomationBench: temperature = 0.6, top_p = 0.95, no max output tokens (bounded by the model's context window).
- BFCLv4: temperature = 0.001, max output tokens = 4096.
- ToolSandbox: temperature = 0, max output tokens = 1024; user simulator: GPT-4o-2024-08-06, temperature = 0, max output tokens = 1024.
- LiveClaw: temperature = 1.0 (vLLM default; the agent sets no temperature), max output tokens = 8192.
- τ³-Bench Banking: official user model (GPT-5.2, low reasoning effort) with default BM25 retrieval, temperature = 0, max output tokens = 16384 per turn.
- WebArena: temperature = 0.6, top_p=0.95, max_turns=40 (no of browser actions allowed), no max output tokens (bounded by the model's context window))
- Other evals: temperature = 0.6, max output tokens = 32768.