{"slug": "lfm", "name": "LFM", "category": "ai-models", "category_name": "AI models", "description": "", "recommenders": 8, "commenters": 9, "warners": 0, "score": 7.035, "rising": 2.69, "first": "2026-01-10", "latest": "2026-10-27", "why": "the LFM2 models run pretty fast on a phone and are surprisingly capable", "why_is_quote": true, "has_verdict": false, "sentiment": {"recommend": 8, "neutral": 1, "compare": 0, "warn": 0}, "spark": [0, 0, 1, 0, 1, 0, 1, 0, 1, 0, 2, 2], "listed": true, "loved": [], "caveat": "", "short_quote": "the LFM2 models run pretty fast on a phone and are surprisingly capable", "kind": "", "icon": true, "domain": "liquid.ai", "github_repo": null, "official_url": "https://www.liquid.ai", "pricing": "", "aliases": ["LFM", "LFM2", "LFM 2.5", "LFM2.5", "LFM3", "LFM 3", "lfm"], "verdict": null, "praise": [{"label": "", "summary": "", "quotes": [{"id": 804214529, "item_id": 50263408, "url": "https://news.ycombinator.com/item?id=50263408", "by": "colingauvin", "date": "2026-10-27", "version": "3", "stance": "recommend", "quote": "Excited to roll out LFM 3 to replace it, looks like a big step up.", "excerpt": "One thing I have been using LFM 2.6 for is annotating emails, text messages, and other metadata. I keep it all ingesting into one giant database and annotate it all as it comes in on an RTX A4500. The LFM annotations are basically triage, and then slowly annotations trickle into my halo model based on lowest confidence (right now that's DeepSeek 4.1 Flash 1017) for an upgraded annotation. (Agents can then search this", "context": "", "use_case": "annotating emails and messages", "switched_from": "LFM 2.6", "story": "Introducing LFM3: Every Modality for Specialized Intelligence"}, {"id": 804211312, "item_id": 50263207, "url": "https://news.ycombinator.com/item?id=50263207", "by": "samstevens", "date": "2026-10-27", "version": "3", "stance": "recommend", "quote": "I think the d1-2B is amazing", "excerpt": "Depends on what you care about. If you want to do local classification/image processing, I think the d1-2B is amazing, it can run on 1GB of unified RAM. Now that it's released and I have weekends off again, I'm going to play with the VL-2B model and the d1-2B and see if I can prompt it to control a small Lego robot arm like this https://bricksrl.github.io/ProjectPage/ If you like agents, I think the 1.7B is very stro", "context": "", "use_case": "Local classification and image models", "switched_from": "", "story": "Introducing LFM3: Every Modality for Specialized Intelligence"}, {"id": 798554993, "item_id": 49909687, "url": "https://news.ycombinator.com/item?id=49909687", "by": "chronicallyoffl", "date": "2026-09-30", "version": "2.5", "stance": "recommend", "quote": "i chose lfm over bert as the base model because it uses grouped-query attention", "excerpt": "hi folks, i'm an engineer and amateur hacker who's been working on a little side project, and i want to show it off and also solicit your feedback. i'm building efficient machine learning models for natural language processing on consumer cpus: sentiment analysis, task routing, document retrieval, etc. i know there are a lot of models for these tasks already, but too many of them are: (1) difficult to configure, or (", "context": "", "use_case": "", "switched_from": "", "story": "Show HN: Offlineisbetter: Efficient Models for Text on CPU"}, {"id": 798461840, "item_id": 49903865, "url": "https://news.ycombinator.com/item?id=49903865", "by": "Baljhin", "date": "2026-09-30", "version": null, "stance": "recommend", "quote": "Tell me your username and i'll give you (six) URLs", "excerpt": "> \"...Tell me your username and i'll give you (six) URLs...\" > (period can be: 7day, 1month, 3month, 6month, 12month or overall) > Recent tracks > Loved tracks > Top tracks > Top Artists > Recommended tracks > Recommended tracks in JSON format https://github.com/xiffy/lfm", "context": "", "use_case": "Last.fm RSS feeds", "switched_from": "", "story": "RSS Feeds for Last.fm"}, {"id": 779950032, "item_id": 48746877, "url": "https://news.ycombinator.com/item?id=48746877", "by": "potus_kushner", "date": "2026-07-01", "version": "2.5", "stance": "recommend", "quote": "these LFM 2.5 models are crazy fast.", "excerpt": "these LFM 2.5 models are crazy fast. the (biggest in series) 8B-A1B model produces 35-40 t/s on an aged 6-core CPU using llama.cpp. it's my go-to model for whenever i need fast local inference. it's also pretty good at toolcalling. would love to see more finetunes on HF, but it appears not many people discovered it yet.", "context": "", "use_case": "fast local inference and toolcalling", "switched_from": "", "story": "Liquid AI releases a 230M model optimized for phones, Raspberry Pi, and robots"}]}], "caveats": [], "who": [], "tags": [], "use_cases": ["annotating emails and messages", "local classification and image models", "last.fm rss feeds", "fast local inference and toolcalling", "edge llm inference"], "compared": [], "makers": [], "threads": [{"url": "https://news.ycombinator.com/item?id=50262229", "title": "Introducing LFM3: Every Modality for Specialized Intelligence", "mentions": 3}, {"url": "https://news.ycombinator.com/item?id=49909686", "title": "Show HN: Offlineisbetter: Efficient Models for Text on CPU", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=49903862", "title": "RSS Feeds for Last.fm", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=48743203", "title": "Liquid AI releases a 230M model optimized for phones, Raspberry Pi, and robots", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=47957648", "title": "LFM2-24B-A2B: Scaling Up the LFM2 Architecture", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=47516650", "title": "Local LLM by Ente", "mentions": 1}], "sentiment_by_month": {"2025-11": {}, "2025-12": {}, "2026-01": {"recommend": 1}, "2026-02": {}, "2026-03": {"recommend": 1, "neutral": 1}, "2026-04": {}, "2026-05": {"recommend": 1}, "2026-06": {}, "2026-07": {"recommend": 1}, "2026-08": {}, "2026-09": {"recommend": 2}, "2026-10": {"recommend": 2}}, "versions": [], "version_unnamed": 3, "family_ref": null, "runs_on": null, "lines": [], "quotes": [{"id": 804240161, "item_id": 50265010, "url": "https://news.ycombinator.com/item?id=50265010", "by": "samstevens", "date": "2026-10-27", "version": null, "stance": "neutral", "quote": "fine-tuned an LFM encoder for promptable classification", "excerpt": "gliner-style models are absolutely crazy, I learned about them a couple weeks ago when a co-worker fine-tuned an LFM encoder for promptable classification. as far as I know, you don't need to fine-tune; you can just prompt for classification. My co-worker's demo is here: https://huggingface.co/spaces/LiquidAI/prompt-routing", "context": "", "use_case": "", "switched_from": "", "story": "Introducing LFM3: Every Modality for Specialized Intelligence"}, {"id": 804214529, "item_id": 50263408, "url": "https://news.ycombinator.com/item?id=50263408", "by": "colingauvin", "date": "2026-10-27", "version": "3", "stance": "recommend", "quote": "Excited to roll out LFM 3 to replace it, looks like a big step up.", "excerpt": "One thing I have been using LFM 2.6 for is annotating emails, text messages, and other metadata. I keep it all ingesting into one giant database and annotate it all as it comes in on an RTX A4500. The LFM annotations are basically triage, and then slowly annotations trickle into my halo model based on lowest confidence (right now that's DeepSeek 4.1 Flash 1017) for an upgraded annotation. (Agents can then search this", "context": "", "use_case": "annotating emails and messages", "switched_from": "LFM 2.6", "story": "Introducing LFM3: Every Modality for Specialized Intelligence"}, {"id": 804211312, "item_id": 50263207, "url": "https://news.ycombinator.com/item?id=50263207", "by": "samstevens", "date": "2026-10-27", "version": "3", "stance": "recommend", "quote": "I think the d1-2B is amazing", "excerpt": "Depends on what you care about. If you want to do local classification/image processing, I think the d1-2B is amazing, it can run on 1GB of unified RAM. Now that it's released and I have weekends off again, I'm going to play with the VL-2B model and the d1-2B and see if I can prompt it to control a small Lego robot arm like this https://bricksrl.github.io/ProjectPage/ If you like agents, I think the 1.7B is very stro", "context": "", "use_case": "Local classification and image models", "switched_from": "", "story": "Introducing LFM3: Every Modality for Specialized Intelligence"}, {"id": 798554993, "item_id": 49909687, "url": "https://news.ycombinator.com/item?id=49909687", "by": "chronicallyoffl", "date": "2026-09-30", "version": "2.5", "stance": "recommend", "quote": "i chose lfm over bert as the base model because it uses grouped-query attention", "excerpt": "hi folks, i'm an engineer and amateur hacker who's been working on a little side project, and i want to show it off and also solicit your feedback. i'm building efficient machine learning models for natural language processing on consumer cpus: sentiment analysis, task routing, document retrieval, etc. i know there are a lot of models for these tasks already, but too many of them are: (1) difficult to configure, or (", "context": "", "use_case": "", "switched_from": "", "story": "Show HN: Offlineisbetter: Efficient Models for Text on CPU"}, {"id": 798461840, "item_id": 49903865, "url": "https://news.ycombinator.com/item?id=49903865", "by": "Baljhin", "date": "2026-09-30", "version": null, "stance": "recommend", "quote": "Tell me your username and i'll give you (six) URLs", "excerpt": "> \"...Tell me your username and i'll give you (six) URLs...\" > (period can be: 7day, 1month, 3month, 6month, 12month or overall) > Recent tracks > Loved tracks > Top tracks > Top Artists > Recommended tracks > Recommended tracks in JSON format https://github.com/xiffy/lfm", "context": "", "use_case": "Last.fm RSS feeds", "switched_from": "", "story": "RSS Feeds for Last.fm"}, {"id": 779950032, "item_id": 48746877, "url": "https://news.ycombinator.com/item?id=48746877", "by": "potus_kushner", "date": "2026-07-01", "version": "2.5", "stance": "recommend", "quote": "these LFM 2.5 models are crazy fast.", "excerpt": "these LFM 2.5 models are crazy fast. the (biggest in series) 8B-A1B model produces 35-40 t/s on an aged 6-core CPU using llama.cpp. it's my go-to model for whenever i need fast local inference. it's also pretty good at toolcalling. would love to see more finetunes on HF, but it appears not many people discovered it yet.", "context": "", "use_case": "fast local inference and toolcalling", "switched_from": "", "story": "Liquid AI releases a 230M model optimized for phones, Raspberry Pi, and robots"}, {"id": 767746192, "item_id": 47984137, "url": "https://news.ycombinator.com/item?id=47984137", "by": "trilogic", "date": "2026-05-02", "version": "2", "stance": "recommend", "quote": "Liquid AI have made some awesome models (especially the smaller ones, they are lightning fast).", "excerpt": "Liquid AI have made some awesome models (especially the smaller ones, they are lightning fast). I wish they made a fast small size coder. Did a finetune distill of 0.8B myself and it is in fact working properly, coding like a 30B model, so I know it is possible. Anyway here you have the 24B parameters with 2B active: https://hugston.com/models/lfm2-24b-a2b-q4-k-m", "context": "", "use_case": "", "switched_from": "", "story": "LFM2-24B-A2B: Scaling Up the LFM2 Architecture"}, {"id": 760377664, "item_id": 47523604, "url": "https://news.ycombinator.com/item?id=47523604", "by": "woadwarrior01", "date": "2026-03-25", "version": null, "stance": "neutral", "quote": "the LFM model license has a commercial use threshold of $10m", "excerpt": "The only takeaway for me from this is that the Ente is at < $10m/year in revenue. Because the LFM model license has a commercial use threshold of $10m. https://www.liquid.ai/lfm-license", "context": "", "use_case": "", "switched_from": "", "story": "Local LLM by Ente"}, {"id": 759875169, "item_id": 47492198, "url": "https://news.ycombinator.com/item?id=47492198", "by": "svachalek", "date": "2026-03-23", "version": "2", "stance": "recommend", "quote": "the LFM2 models run pretty fast on a phone and are surprisingly capable", "excerpt": "A long time. But check out Apollo from Liquid AI, the LFM2 models run pretty fast on a phone and are surprisingly capable. Not as a knowledge database but to help process search results, solve math problems, stuff like that.", "context": "", "use_case": "", "switched_from": "", "story": "iPhone 17 Pro Demonstrated Running a 400B LLM"}, {"id": 745059249, "item_id": 46566203, "url": "https://news.ycombinator.com/item?id=46566203", "by": "kristopolous", "date": "2026-01-10", "version": "2.5", "stance": "recommend", "quote": "for edge running look into the LFM 2.5 class of models", "excerpt": "I've been working on a more unixy version of his tool I call llcat https://github.com/day50-dev/llcat It might help things get closer. Also for edge running look into the LFM 2.5 class of models: https://huggingface.co/LiquidAI/LFM2.5-1.2B-Instruct", "context": "", "use_case": "edge LLM inference", "switched_from": "", "story": "Ask HN: What's a standard way for apps to request text completion as a service?"}]}