--- title: Run fast, private agents on-device with LFM2.5-24B-A2B description: Liquid AI released LFM2.5-24B-A2B, a 24B-parameter mixture-of-experts reasoning model with only 2B active parameters per token, designed for private, on-device... canonical: https://daily.dev/posts/run-fast-private-agents-on-device-with-lfm2-5-24b-a2b-r5wlevpo4 twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Run fast, private agents on-device with LFM2.5-24B-A2B | daily.dev og:description: Liquid AI released LFM2.5-24B-A2B, a 24B-parameter mixture-of-experts reasoning model with only 2B active parameters per token, designed for private, on-device... og:url: https://daily.dev/posts/run-fast-private-agents-on-device-with-lfm2-5-24b-a2b-r5wlevpo4 og:image: https://api.daily.dev/og/posts/R5WLevPo4.png og:image:alt: Run fast, private agents on-device with LFM2.5-24B-A2B og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Run fast, private agents on-device with LFM2.5-24B-A2B **[Hugging Face](https://daily.dev/sources/huggingface)** · 5 min read · 0 upvotes · 0 comments ## Summary Liquid AI released LFM2.5-24B-A2B, a 24B-parameter mixture-of-experts reasoning model with only 2B active parameters per token, designed for private, on-device agents. It builds on LFM2-24B-A2B, extends context to 131K tokens, expands the tokenizer vocabulary to 128K, and was trained on ~33T pre-training tokens plus agentic reinforcement learning inside harnesses like OpenClaw, Hermes Agent, Pi, and OpenCode. Benchmarks show it competitive with or ahead of larger MoEs (Qwen3.6-35B-A3B, Gemma-4-26B-A4B-it, Nemotron-3.5-Lightning) on tool-use and instruction-following tasks, while trailing on knowledge and coding benchmarks. It reaches 184 tok/s on an Apple M5 Max and over 21K output tok/s on a single NVIDIA B300 GPU, and ships with day-one support in llama.cpp, MLX, vLLM, SGLang, and ONNX. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### What is LFM2.5-24B-A2B and how does it compare to LFM2-24B-A2B? LFM2.5-24B-A2B is a 24B-parameter mixture-of-experts reasoning model with 2B active parameters per token, built by Liquid AI for on-device agents. Unlike its predecessor LFM2-24B-A2B, it is reasoning-only, producing an explicit chain of thought before answering. Context grows from 32K to 131K tokens, vocabulary doubles from 65K to 128K tokens, and pre-training scales from ~17T to ~33T tokens. _Developers tracking new MoE model releases for local agent stacks can follow updates like this on daily.dev._ ### How fast is LFM2.5-24B-A2B on consumer hardware compared to GPU servers? LFM2.5-24B-A2B decodes 184 tokens/s on an Apple M5 Max and 114 tokens/s on a Ryzen AI Max+ 395, up to 1.7x faster than the next-best comparable model while using the least memory. On a single NVIDIA B300 GPU, it reaches over 21,000 output tokens per second at high concurrency, making it the fastest model in its size class. _Anyone benchmarking on-device versus GPU inference speeds can keep tabs on comparisons like this via daily.dev._ ### How does LFM2.5-24B-A2B perform against Qwen3.6-35B-A3B on agentic benchmarks? LFM2.5-24B-A2B leads or matches Qwen3.6-35B-A3B on five of eight tool-use and agentic benchmarks despite having roughly 50% fewer total and active parameters. It scores 48.09 versus 38.19 on LiveClaw and over three times the next-best model's score on AutomationBench (27.69 vs 8.72), though Qwen3.6-35B-A3B leads on WebArena Verified, GPQA Diamond, and SWE-Bench Verified. _Teams deciding between smaller and larger MoE models for agent tool use can track benchmark trade-offs like these on daily.dev._ ## Similar posts on daily.dev - [Needle 2 - The 14 MB Agentic LLM for Tiny Devices](https://daily.dev/posts/needle-2---the-14-mb-agentic-llm-for-tiny-devices-eu1fgqsim) · Hacker News · 4 upvotes · 0 comments --- Tags: [#ai-agents](https://daily.dev/tags/ai-agents), [#mixture-of-experts](https://daily.dev/tags/mixture-of-experts) [View this post on daily.dev](https://daily.dev/posts/run-fast-private-agents-on-device-with-lfm2-5-24b-a2b-r5wlevpo4) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Run fast, private agents on-device with LFM2.5-24B-A2B","url":"https://daily.dev/posts/run-fast-private-agents-on-device-with-lfm2-5-24b-a2b-r5wlevpo4","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/run-fast-private-agents-on-device-with-lfm2-5-24b-a2b-r5wlevpo4"},"datePublished":"2026-10-13T17:09:19.233Z","dateModified":"2026-10-14T03:51:25.526Z","description":"Liquid AI released LFM2.5-24B-A2B, a 24B-parameter mixture-of-experts reasoning model with only 2B active parameters per token, designed for private, on-device...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/4ed73b588b7d90f5427c6aeed7e6f1e9?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/4ed73b588b7d90f5427c6aeed7e6f1e9?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Hugging Face","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Hugging Face","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/f1f55c67d81a4330acf5b90b26b0c8e1","url":"https://daily.dev/sources/huggingface"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/run-fast-private-agents-on-device-with-lfm2-5-24b-a2b-r5wlevpo4","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai-agents,mixture-of-experts","timeRequired":"PT5M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Hugging Face","item":"https://daily.dev/sources/huggingface"},{"@type":"ListItem","position":3,"name":"Run fast, private agents on-device with LFM2.5-24B-A2B"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/run-fast-private-agents-on-device-with-lfm2-5-24b-a2b-r5wlevpo4#faq","mainEntity":[{"@type":"Question","name":"What is LFM2.5-24B-A2B and how does it compare to LFM2-24B-A2B?","acceptedAnswer":{"@type":"Answer","text":"LFM2.5-24B-A2B is a 24B-parameter mixture-of-experts reasoning model with 2B active parameters per token, built by Liquid AI for on-device agents. Unlike its predecessor LFM2-24B-A2B, it is reasoning-only, producing an explicit chain of thought before answering. Context grows from 32K to 131K tokens, vocabulary doubles from 65K to 128K tokens, and pre-training scales from ~17T to ~33T tokens. Developers tracking new MoE model releases for local agent stacks can follow updates like this on daily.dev."}},{"@type":"Question","name":"How fast is LFM2.5-24B-A2B on consumer hardware compared to GPU servers?","acceptedAnswer":{"@type":"Answer","text":"LFM2.5-24B-A2B decodes 184 tokens/s on an Apple M5 Max and 114 tokens/s on a Ryzen AI Max+ 395, up to 1.7x faster than the next-best comparable model while using the least memory. On a single NVIDIA B300 GPU, it reaches over 21,000 output tokens per second at high concurrency, making it the fastest model in its size class. Anyone benchmarking on-device versus GPU inference speeds can keep tabs on comparisons like this via daily.dev."}},{"@type":"Question","name":"How does LFM2.5-24B-A2B perform against Qwen3.6-35B-A3B on agentic benchmarks?","acceptedAnswer":{"@type":"Answer","text":"LFM2.5-24B-A2B leads or matches Qwen3.6-35B-A3B on five of eight tool-use and agentic benchmarks despite having roughly 50% fewer total and active parameters. It scores 48.09 versus 38.19 on LiveClaw and over three times the next-best model's score on AutomationBench (27.69 vs 8.72), though Qwen3.6-35B-A3B leads on WebArena Verified, GPQA Diamond, and SWE-Bench Verified. Teams deciding between smaller and larger MoE models for agent tool use can track benchmark trade-offs like these on daily.dev."}}]} ```