From the community. See what people are saying.

okay no longer trapped inside today :> i started running a shitty vibe coded agent on my jetson orin nano called laine :D (yes from SEL). it’s using LFM2 8B A1B which so far is an amazing model so far. I love it so far

AI that runs offline on your phone. LFM2.5-1.2B-Thinking uses under 1GB, CPU only, 10 to 15 tokens per second, about 1.5 second latency. Scores 72% GSM8K, 68% ARC. Apache 2.0 on Hugging Face Jan 20 2026. Faster and lighter than Phi-2 and Gemma 2B, but clouds still lead.

@TheAhmadOsman These are big names the real benificial models are the ones that are bringing the cost of intelehence down namely the 1-10 b prams pioneerd by the likes of @liquidai and @GeminiApp with gemma with their amazing 1 b prams model which can run inside your mobile phones natively

@AntDX316 The response quality of LiquidAI/LFM2.5-1.2B-Thinking and LFM2.5-1.2B-Instruct is very high, which is quite remarkable for such small models.

LFM2.5-1.2B-Thinking and Instruct lightning speed ...of the model have also a 1.6b vision model which can process images quite accurately. It was tested in cpu/gpu and flash attention with a max speed in one of our servers of 342 tokens per second. Definitely worth using and having in the repo...

@liquidai Impressive benchmarks for LFM2.5-1.2B-Thinking, especially with on-device capabilities and concise reasoning - a significant step forward in AI efficiency.

Great leap in on-device AI: Liquid's new LFM2.5-1.2B-Thinking does genuine step-by-step reasoning using just ~900 MB RAM, runs on basically any modern phone. Beats larger models like Qwen3-1.7B on math/tool use while being dramatically faster & leaner. Privacy + zero-latency

@songdng @liquidai Just tried it peak memory usage of barely around 2.5Gb for a thinking model is crazy good

Emerging AI tools like Liquid AI’s compact reasoning model running on smartphones show how decentralized, real-time intelligence is becoming more accessible for market insights on the go. https://huggingface.co/LiquidAI/LFM2.5-1.2B-Thinking

Nice! I will test it today, the instruct version punches way above its weight, but I usually don't get good results with small thinking models because they enter in a thinking loop, but it seems there was a focus on preventing that. Also, there is a mention saying the model is not suitable for coding, do you plan to release a coding capable (even if not code focused) in the future? The previous 8b moe had additional training tokens of code. With the tool call capabilities of lfm + small memory foot print of context length, a code capable lfm2.5 8b moe would be amazing

Liquid AI LFM 2.5 1.2B Thinking Model Outperforms Larger Models on Reasoning Benchmarks Liquid AI just dropped the LFM-2.5-1.2B-Thinking, a tiny 1.2 billion parameter model designed for fast, private reasoning right on your device. It runs completely offline and uses less than 900MB of memory, perfect for phones and edge devices without any drop in capability. What sets it apart is its clean, straight-to-the-point reasoning traces, blazing-fast inference, and excellent performance on instruction following, tool use, and math problems.Benchmarks show it beating much bigger models like Qwen3-17B in thinking mode on tests such as GPQA Diamond (37.86%), MMLU-Pro (49.65%), and....

Another great model from Liquid AI LFM2.5-1.2B-Thinking. What needed a data center 2 years ago now runs fully on-device—on any phone with \~900MB RAM. Concise reasoning. Thinks before answering. Edge-scale latency. Tools, math, instructions—nailed

@maximelabonne @huggingface Congrats! I pre-trained a model using the LFM2 architecture a couple days ago, it's very efficient for training. Forgive me if I'm mistaken, why haven't you guys put out any research directly comparing the LTC architecture with hybrid-attention or regular transformers in LLMs?

It is incredible, hard to believe a 1B model can sound coherent, but you can straight up have conversations with it. Got me excited for Gemma4 and the future of smaller models

What's better than an AI LLM chatbot? 10 agents with different specialties and personas responding to you realtime like the boss! (note: video not sped up + autocomplete in bg). Impressed with @liquidai LFM2.5 1.2B model (@UnslothAI version) running all locally with @lmstudio

@paulabartabajo_ @liquidai This is impressive. A 1.2B model achieving 239 tok/s on an AMD CPU and 82 tok/s on a mobile NPU while using under 1GB RAM makes truly private, offline, production-grade inference on everyday devices realistic now.

@maximelabonne Liquid LFM2.5 is incredible, the fact you are able to do so much with 1.2B is absurd!

Liquid AI's new models are changing the game! Faster, more efficient generative AI for any device. The future is here! #AI #MachineLearning #Innovation #Tech #FutureTech #LiquidAI

@liquidai is the best model I’ve worked with for post-training and fine-tuning, especially at 2–3B params. Most open-source models (Qwen, DeepSeek, etc.) feel over-tuned and hard to adapt.

@maximelabonne @Lightricks Seeing Liquid AI models trending is a strong signal for non-Transformer architectures! 💧 At 1.2B params, these are perfect for experimenting with efficient local inference. Great to see diversity in the open-source leaderboard! 🚀 #AI #HuggingFace

An AI model optimized for on-device deployment with efficient performance and low memory usage: @liquidai https://huggingface.co/LiquidAI/LFM2.5-1.2B-Instruct

The latest @liquidai's LFM2.5 1.2B model is really impressive, my favorite 1B model right now. It’s relatively smart, lightweight and super fast running on iPhone with MLX. Go try it if you haven’t!

@its_maddy_a @liquidai @ZeroGPU_AI This is awesome to hear! 🚀 Liquid AI’s architecture is a game-changer for efficiency. We’d love to see how @ZeroGPU_AI pushes the limits of the LEAP SDK. Keep us posted on the progress!

🚨 @liquidai just dropped LFM<2.5-1.2B-Instruct 🎆♠ and the architecture is actually insane 🚀 🔹 Hybrid design: Convolutions + Attention (not pure transformer) 🔹 1.2B params but rivals 3B+ models 🔹 239 tok/s on AMD CPU (runs under 1GB RAM) 🔹 Built for agentic workflows

Damn you guys keep knocking it out of the park. Congratulations on all the amazing releases. Im still trying to fully utilize LFM2-1.2B then you drop new weights. So fun to be here and experience the fruits of your toil. Thanks for looking out for those of us with less computational capacities. Its amazing how far small models have come and you're an integral part of that puzzle. Best wishes

@Prince_Canuma @liquidai LFM2.5-Audio's end-to-end processing is brilliant-especially those latency cuts! MLX-Audio integration could redefine on-device voice tech. 🚀

@kimmonismus I tried this on my phone and it's a really good model. I enjoyed using it.

The future isn't in the Cloud, it's in your pocket. We obsess over 100B parameter models, but the real revolution is happening at the micro-scale. LFM 2.5 proves architecture beats raw size. With just 1.2B parameters, it crushes Llama 3.2 1B on complex reasoning, scoring

LFM2.5-1.2B-Instruct from LiquidAI works well with opencode and it's blazing fast

@liquidai cool release. what's the secret sauce for sub-1b latency on-device agents? custom kv cache compression or fused ops? broader modalities – audio+vision in 1b? digging the agentic focus.

@liquidai Qwen3 4B 2507 Instruct might be an unfair comparison cause it is 2x the size, but thats what I daily drive. When I compare LFM against Qwen3 1.7B or Granite, LFM absolutely crushes the others and it is not even close. LFM2.5 is the best <4B model currently out there.

@liquidai LFM2.5 1.2 reminds me of LFM2 700M, it's just more powerful... great job, your models are really excellent.

🔒 Game changer: AMD Ryzen AI runs Liquid AI's LFM2 model locally for meeting summaries. Zero internet needed, complete privacy, 16-second processing. Finally, AI PCs that don't send your data to the cloud. #AdwaitX #AMD #PrivacyFirst #AI #TechNews #CES2026 #AINews #AIComputing #EnterpriseAI

This is pretty interesting, especially the focus on "real on-device agents". Curious if anyone has benchmarked tool-use style workflows (planner/executor, function calling, etc) on LFM2.5 yet, and how it compares to small Llama variants when latency really matters. If youre exploring agentic patterns, Ive been collecting some practical notes and examples here too: https://www.agentixlabs.com/blog/

Liquid AI’s LFM2.5 packs multimodal, agentic AI into sub-2B models that run entirely on-device—text, vision, and real-time speech, no cloud needed. But small models still can’t match frontier-scale reasoning; they excel at speed, privacy, and efficiency, not complexity. Already

1. Exciting AI breakthroughs are here! New compact models like LFM2.5 bring powerful on-device intelligence with billions of parameters and smarter multimodal abilities. Tencent’s Youtu-LLM outperforms bigger models in coding & math tasks. 🤖✨

Holy moly, Liquid AI just unveiled LFM2.5, a powerful open-weight model family designed to run fast, private, and always-on directly on devices. LFM2.5 sets a new bar for edge AI across text, vision, audio, and Japanese-language use cases.The TTS examples are amazing btw

Stop everything and grab this Vision Language Model fast! 📱👁️ This tiny 3B model is a breakthrough in Edge AI and Computer Vision. It’s faster than the giants and runs perfectly on mobile.  #liquidai #edgeai #computervision #tinymodel #opensource #aitechnology

I got frustrated dealing with massive responses from many MCPs and threw something together over the last couple days... it might help you too. Or not!...you're playing with multiple AI tools/coding assistants and hate having to reconfigure MCPs for each one- Very configurable options to override behavior globally or different tools via a single JSON file, plus a UI for management and visibility. I've been testing with a high-quant Qwen3-0.6b and LFM2-1.2b and it's doing very well for me. For example, I have it use web search and fetch for URLs and instead of having the larger model process the entire pages, the tiny model reads the page up to 10x faster, and just gives the large model the answers it needs, also keeping context lower.

📰 LFM2 2.6B-Exp on Android: 40+ TPS and 32K context. The article discusses the impressive performance of LiquidAI's new LFM2 2.6B-Exp model, which can achieve GPT-4 level performance across a wide range of benchm… Read more 👇https://ai-curator.jp/articles/cmjw8efiq00v2ct1epceqo9q1 #AI #AINews

Goodbye to 2025 👋Happy New Year 2026! Small LLMs highlights in 2025: Qwen3-4B; LFM2–8B-A1B / LFM2-2.6B-Exp; granite-4.0-h-tiny (7B A1B); gemma-3n-E4B / gemma-3-4b-it; SmolLM3-3B; Ministral-3-3B. They run at decent speeds on the CPU! Looking forward to 2026's smol LLMs!

@Henry_Ndubuaku @liquidai 231MB peak for a 1.6B INT8 VLM across context sizes is wild. Curious what dominated after quantization, KV cache, tokenizer, or runtime overhead? Also love the “budget devices first” stance.

LFM2-2.6B-Exp shows pure RL beating giants: a 3B model tops a 263x larger peer on IFBench, with think tokens and on device tool use for edge reasoning. Is edge AI finally reality? #EdgeAI

Liquid AI released an experimental checkpoint of LFM2-2.6B using pure RL, making it the strongest 3B on the market. Meet the strongest 3B model on the market. LFM2-2.6B-Exp is an experimental checkpoint built on LFM2-2.6B using pure reinforcement learning. Consistent

@LocallyAIApp @MistralAI The new LFM2-2.6B-EXP is incredible for local models and runs perfectly. Would it ever arrive on LOCALLYAI?

The shift from bigger-is-better to efficient-is-smarter is accelerating. Liquid AI's 3B model beating much larger ones proves that pure RL training can unlock incredible efficiency gains.

Amazed by LFM 3b model.. and amazed by the @liquidai team for making this amazing model within 3 b parameter... running quantized Q4 but already loving it... earlier i was running gemma2 but looks like LFM is my new love...

Just tried LFM2 2.6 EXP model and it's mind-blowing to me. It's my go-to model for my phone with ChatterUI 😁 https://huggingface.co/LiquidAI/LFM2-2.6B-Exp

Liquid AI released an experimental checkpoint of LFM2-2.6B using pure RL, making it the strongest 3B on the market. "Meet the strongest 3B model on the market. LFM2-2.6B-Exp is an experimental checkpoint built on LFM2-2.6B using pure reinforcement learning. Consistent improvements in instruction following, knowledge, and math benchmarks. Outperforms other 3B models in these domains. Its IFBench score surpasses DeepSeek R1-0528, a model 263x larger"

@liquidai Incredible performance! Congratulations @liquidai !

@maximelabonne The real breakthrough here isn't just the 2.6B size; it's the proof that Liquid Foundation Models (LFMs) respond to pure RL just as well as (or better than) traditional Transformers. For a long time, the 'Reasoning' meta was locked into the Attention mechanism. Seeing...

@liquidai Impressive results for a 3B model pure RL is paying off.

Small but Mighty Liquid AI's LMF2-2.6B-Exp punching much above its weight

🚀 LiquidAI just released LFM2‑2.6B‑Exp, a 3 B‑parameter text‑gen model that scored 683 on the latest benchmark — updated just 3 hrs ago. It brings near‑GPT‑4 quality with a tiny footprint. https://huggingface.co/LiquidAI/LFM2-2.6B-Exp

Small models are getting scary good. Liquid AI shows how reinforcement learning can turn a 2.6B model into something that beats most 3B LLMs — perfect for edge, local and real-world deployments. #AI #LLM #EdgeAI #MachineLearning #ReinforcementLearning

@maximelabonne Interesting checkpoint. LFM2-2.6B-Exp seems to leverage pure RL effectively for instruction following and reasoning, outperforming most 3B class models in specialized tasks.

Wow! @liquidai just released a 2.6B model that beats every 3B on almost all benchmarks! LFMs are becoming the best edge-device AI models! Though not surprised, amazing team!

GPT 4 level AI with 3B parameters is insane! Absolutely going to give it a try. ‍

Liquid AI's LFM2-2.6B-Exp got 42% in GPQA - incredible for a 2.6B param mode. This level of score would need a much larger model, but LFM2-2.6B is doing it mostly by changing the training signal, not the architecture - just by adding RL on top of the same base checkpoint.

@liquidai Whoa, a 3B model outperforming one 263× its size? This is next-level RL magic! ‍

🤯 🎁 Liquid AI brings "Deep Thinking" to a 2.6B model / your Phone / your robot 🤖Liquid AI released a holiday present for your new robot 🎁 They just cracked the code on efficient intelligence. With LFM 2-2.6B-Exp, the era of bloated, memory-hogging Transformers running on

This is actually nuts: 41.3% on GPQA for a 2.6B model! That is PhD level knowledge running locally on your iPhone. The efficiency gains right now are insane.

@liquidai The leap in instruction-following is impressive for a 3B model. How do you see models like this shifting enterprise or R&D applications?

Local LLM models are getting good. LFM2-2.6B-Exp is quite powerful and fast on Apple Silicon.

For ~3B model size, its instruction-following and reasoning results are genuinely impressive.

Small AI designed to follow instructions, run locally and is totally open weights. LFM2-2.6B-Exp technical breakdown: F32 tensor; 2.6B params; For edge devices; Agent & RAG focused; Math skills (AIME25); Good general knowledge (GPQA); Strong prompt adherence (IFBench)

1. The core breakthrough: LFM2 models run 2x faster on regular CPUs compared to similar models while maintaining impressive capabilities. No more choosing between performance and accessibility.

@liquidai Bottom line:LFM2 models show that you can get 3B–4B-level quality from ~1B–2.6B models and run them 2–3× faster on a phone. Small models, if designed correctly, can dominate the accuracy–latency tradeoff.

One of the things im going to do this weekend is run BFCL on LFM2 and Arcee's new tiny MOE Trinity Nano; for me tool calling is .... the benchmark, lets say. That and context i guess. So i'll post about that at some point. But so far, anecdotally just playing with it LFM2 8B A1B is still impressive.

#LiquidAI, an @MIT spinoff, just released a blueprint for enterprise-grade small models. Small can be mighty: high performance, low latency, on-device #AI. 51-page roadmap now public. 🚀#EnterpriseAI #OnDeviceAI @VentureBeat @carlfranzen @liquidai

What if BabyAGI didn’t need the cloud? We took @yoheinakajima's BabyAGI loop and ran it locally on an iPhone over a small language model (LFM2-350M). No OpenAI key. No servers. Just @RunAnywhereAI SDK + @liquidai LFM2-350M running everything fully on-device.

@MParakhin Wow, Mikhail, that's some impressive performance! Sub-20ms inference is quite something, and the architecture sounds game-changing, right?

Two amazing companies, now more amazing with this partnership!You're going to start hearing a lot more about @LiquidAI_ (a @PWVentures investment) in the future; their novel AI architecture blows away transformers in a ton of valuable use cases.Congrats to both sides on

@MParakhin Honestly, that’s insane sub 20ms inference with fewer parameters and higher performance? Liquid’s LFMs are setting a new bar for real-world AI deployment.

@ShopifyEng @LiquidAI_ Game changer for e-commerce! Sub-20ms models and a fresh recommender will definitely enhance user experience. Excited to see the results!

@LiquidAI_ @Shopify Impressive collaboration — sub-20ms inference and measurable conversion gains show how foundational models can deliver real, production-grade impact in commerce. #AI #TechAI

Liquid AI launched LFM2-VL-3B, a compact multilingual vision model supporting ten languages, excelling in single- and multi-image understanding and English OCR, scoring 51.8% on MM-IFEval and 71.4% on RealWorldQA with notably low hallucination rates.

@LiquidAI_ LiquidAI/LFM2-8B-A1B model is really fast on iPhone. I must say I really like the vibe of this model for an on-device model. Any other LLMs I can check out?

5️⃣ Jan Nano An underrated agent model. Executes, plans, automates. Feels like a real assistant. 6️⃣ LFM2 VL 1.6BCompact multimodal engine. Fast inference. Handy for on device vision tasks. 7️⃣ Magistral Small 1.2 Solid visual plus language mix. Consistent and easy to use.

@alshell7 @LiquidAI_ There are also the currently available LFM2-1.2B-Extract and LFM2-350M-Extract. Not tuned for PII but they will do the job very well. Been trying them in the last couple days for a few things and have to say, they are amazing models.

Hermes 4 14B: Completely uncensored and answers all the questions that other LLMs reject. Jan-Nano: Excellent agentic model for using tools and automating many tasks LFM2-VL 1.6B: Light weight and extremely fast multimodal model with vision

Liquid AI has unveiled the LFM2-8B-A1B, a groundbreaking Mixture-of-Experts (MoE) model designed for on-device execution. With an impressive

@LiquidAI_ Congratulations on the launch of LFM2-8B-A1B! Achieving performance comparable to larger models while significantly enhancing inference speed is a remarkable feat. The efficiency of on-device MoE is crucial for real-time applications. I'm excited to see how this innovation will

@maximelabonne @huggingface Impressive specs on LFM2-8B-A1B. The MoE architecture could change the game for mobile usage. Excited to see those benchmark results!

...and another one to the MIX 🔥 LFM2-8B-A1B takes MoE on-device: fast, private, and seriously efficient. Big-model brains, phone-level speed

@pasashque @Apolloaiapp @LiquidAI_ We use LEAP Edge SDK to work with Liquid models. U can check it here ⬇️https://docs.liquid.ai/leap/edge-sdk/android/android-quick-start-guide

Cool release by @LiquidAI_: LFM2-Audio-1.5B It’s a pretty cool omni-architecture that enables prediction of both text and audio tokens, meaning it can handle multi-turn S2S, ASR, and TTS (with voice description) within a single model. Great to see, once again this year, a model

Hell yes! Currently running a 2.6B liquid foundation model on my phone. Imagine how useful this could be in any number of scenarios without network access!

In my limited experience testing various nano/tiny/small models from various labs, LFM2-1.2B was galaxies ahead of competitors. It's 1.2B but really behaves more like a ~10B model in terms of coherence and reasoning power. Even their smaller ~300M model feels more like a 4B, which is mind blowing because it runs fast even on a CPU (hell, it runs fast even on a browser via WebAssembly, unfathomably amazing). I'm hoping that these specialized models will be very good, I have a good amount of trust in Liquid AI.

Game-changing on-device AI just dropped. 🤯Liquid AI unleashed LFM2-Audio, a 1.5B param speech model with a crazy low 95ms latency—that's literally the speed of a blink. This is huge. Try it here: https://playground.liquid.ai/login?callbackUrl=%2FtalkBut that's not all... 🧵

🚨 GAME CHANGER ALERT: MIT just broke the AI game! 🤯 Liquid AI's new models are DESTROYING traditional LLMs and here's why this is HUGE 👇✨ Uses 90% FEWER neurons but performs BETTER 🚀 Handles 1 MILLION tokens with minimal memory 💡 Perfect for edge devices (your phone could run this!) 🔥 LFM-1B is setting NEW records on every benchmark. While everyone's obsessing over bigger models, Liquid AI said "hold my beer" and went LIQUID 💧 This isn't just an upgrade - it's a complete paradigm shift. Traditional transformers are about to look like flip phones 📱➡️🧠The future of AI just got a lot more interesting... and efficient! 🌟 What do you think? Are liquid neural networks the future? Drop your thoughts below! 👇#LiquidAI #AI #MachineLearning #TechNews #Innovation #MIT #NeuralNetworks #ArtificialIntelligence #TechBreakthrough #FutureOfAI

Liquid AI just dropped something that kills the entire whisper->gpt->elevenlabs pipeline. sub-100ms end-to-end voice with unified audio i/o - no more stitching 3 apis together and dealing with streaming headaches. this changes everything for voice apps

LiquidAI is the real deal. This company will catch up quick. Their 40B LFM is cool as hell.

LiquidAI bet on small but mighty model LFM2-1.2B-Tool/RAG/Extract. So LiquidAI just announced their fine-tuned LFM models with different variants - Tool, RAG, and Extract. Each one's built for specific tasks instead of trying to do everything. This lines up perfectly with that Nvidia whitepaper about how small specialized models are the future of agentic AI. Looks like it's actually happening now. I'm planning to swap out parts of my current agentic workflow to test these out. Right now I'm running Qwen3-4B for background tasks and Qwen3-235B for answer generation. Gonna try replacing the background task layer with these LFM models since my main use cases are extraction and RAG. Will report back with results once I've tested them out.

The rise of Liquid AI is worth watching 👀. Built for efficiency, scalability & speed — it’s redefining how AI systems learn and adapt. Feels like we’re just scratching the surface of what this tech can unlock. 🚀 #AI #LiquidAI #DeepTech

Very good little model released quietly. In testing it's quite competent and very fast. Quants available on HF.

@LisaFlorentina8 Liquid AI is the future for sure

🚀 Running Small LMs on Mobile (and in the Browser)I’ve been testing small language models (<1B params) on a Samsung tablet (8 GB RAM). Results are solid and show how far on-device AI has come.

LiquidAI is unleashing new real-time translation power! 🤯 Introducing LFM2-350M, a model fine-tuned for Japanese-English communication! This is a total game-changer for breaking language barriers! 🚀✨

Liquid AI has open-sourced two compact vision-language models, LFM2-VL-450M (350 M LM + 86 M vision) and LFM2-VL-1.6B (1.2 B LM + 400 M vision). Both deliver near-real-time inference that’s about 2× faster on GPUs than many existing VLMs while preserving competitive accuracy.

Exciting news from Liquid AI! 🚀 Introducing LFM2-VL: super-fast, open-weight vision-language models perfect for low-latency, on-device deployment. Revolutionizing AI for smartphones, laptops, wearables, and more! #AI #VisionLanguageModels

🚀 @LiquidAI_ LFM2-VL-1.6B sets new benchmarks: super-fast, edge-ready vision-language AI for real-world applications. Efficient. Scalable. Open. Explore now 👉 https://huggingface.co/LiquidAI/LFM2-VL-1.6B

450M vision model runs at 353 tokens/sec on MLX using just 1.456GB memory 🔥. LiquidAI drops LFM2-VL-450M → MLX-VLM community with @Prince_Canuma ports it → Your Mac becomes a vision AI beast. Liquid neurons adapting computation on the fly. 450M params punching way above its weight class. Open source wins again. Model drops, community delivers, boom - it's local. What are you building with instant vision AI?

Liquid AI announced LFM2-VL, fast and lightweight vision models (450M & 1.6B). 2 models based on the hybrid LFM2 architecture: LFM2-VL-450M and LFM2-VL-1.6B. Available quant: 8bit MLX, GGUF Q8 & Q4 (llama.cpp release b6183)

Liquid AI just launched a vision model so light it can run on your smartphone, big leap for AI everywhere. No more cloud dependency—privacy and speed go next-level. Ready to build the future? Check the VentureBeat article. Daily tips -> @realjackhui #AIonMobile

🔥 A 1.2B translation model that punches way above its weight. SauerkrautLM-Translator-LFM2.5-1.2B just dropped. This isn't another general-purpose LLM. It's a specialized translator built for one thing — high-quality, nuanced text conversion. What makes..

@localghost Big congrats, on device and Liquid AI feels like a game changer combo.

Smallest Model (600mb) I've tried that knows the legend of the The Lost R -- Book 1 of the Strawberry Saga and 3 Rs trilogySO FAST https://lmstudio.ai/models/liquid/lfm2-1.2b

Trying out LFM2 350M from @LiquidAI_ and was mind-blown 🤯 The responses were very coherent. Less hallucinations compared to models of the same size. Very well done!! The best part: Q4_K_M quantization is just 230 Megabytes, wow!

Liquid AI Open-Sources LFM2: A New Generation of Edge LLMs. Liquid AI just dropped a game-changer for edge computing with LFM2, their second-generation foundation models that run directly on your device. These aren't just incremental improvements—we're talking 2x faster inference

LFM2 GGUF models just dropped for llama.cpp - these are game changers. 2x faster than Qwen3 on CPU, 200% higher throughput vs competitors, designed for edge deployment. Finally, high-performance models that actually run well locally: https://huggingface.co/LiquidAI/LFM2-1.2B-GGUF

LFM2-1.2b is very very usable on my pathetic 8GB M2 MacBook Air. Love it.