Best models for your hardware this week. 8-12GB - https://huggingface.co/LiquidAI/LFM2.5-8B-A1B… incredible model, so fast, so small 16-32GB - latest Google model, Gemma 12B: https://huggingface.co/google/gemma-4-12B… really solid performance up neck and neck with a model 2x its size from a month ago. Jetbrains new model, best in class on livecode bench 32-96gb - Nex-N2-Mini GPT style postrain of Qwen-35B it seems to be its class leader caveman style reasoning https://huggingface.co/nex-agi/Nex-N2-mini… - Jackrong’s Qwopus is the #1 overall Q4 of Qwen3.6-27B on our benchmark suite of 5 agent + coding benchmarks (1200 samples total) https://huggingface.co/Jackrong/Qwopus3.6-27B-v2-MTP-GGUF… 192gb - Step-3.7-Flash is hard to beat, high scores, really fast inference, vision capable, later cutoff dates https://huggingface.co/stepfun-ai/Step-3.7-Flash… 384gb - Nex-N2-Pro GPT style post train of Qwen-3.5-397B incredibly strong and #1 on deepswe if their claims are right https://huggingface.co/nex-agi/Nex-N2-Pro… 768gb - very promising post-train of GLM-5.1 that wins out on 8 benchmarks https://huggingface.co/mindlab-research/Macaron-V1-Preview-749B…

LiquidAI/LFM2.5-8B-A1B · Hugging Face

Source: https://huggingface.co/LiquidAI/LFM2.5-8B-A1B

**⚠️Important:**The tokenizer was updated after the original release to fix tool-calling issues inllama\.cpp. If you downloaded LFM2.5-8B-A1B beforecommitfeb5e04, please re-download the tokenizer files. TheGGUF fileshave also been re-converted with the updated tokenizer.

LFM2.5 is a new family of hybrid models designed for on-device deployment. It builds on the LFM2 architecture with extended pre-training and reinforcement learning.

• On-device personal assistant: Designed to power real-life applications, chaining tool calls, and following complex instructions on all devices.
• Compressed performance: Competitive with much larger dense and MoE models on instruction following and agentic tasks.
• Unmatched throughput: Fastest in its size class on both CPU and GPU inference, with day-one support for llama.cpp, MLX, vLLM, and SGLang.

Find more information about LFM2.5-8B-A1B in ourblog post.

*AA-Omniscience Index (higher is better) rewards correct answers and penalizes hallucinations. Scores range from -100 to 100. See more results onArtificial Analysis.

https://huggingface.co/LiquidAI/LFM2.5-8B-A1B#%F0%9F%97%92%EF%B8%8F-model-details🗒️ Model Details

ModelParametersDescriptionLFM2.5-8B-A1B-Base8.3B total / 1.5B activePre-trained base model for fine-tuningLFM2.5-8B-A1B8.3B total / 1.5B activeReasoning-tuned general-purpose model LFM2.5-8B-A1B is a general-purpose text-only model with the following features:

• Total parameters: 8.3B
• Active parameters: 1.5B
• Number of layers: 24 (18 double-gated LIV conv + 6 GQA)
• Training budget: 38 trillion tokens
• Context length: 128,000
• Vocabulary size: 128,000
• Languages: English, Arabic, Chinese, French, German, Italian, Japanese, Korean, Portuguese, Spanish
• Generation parameters: We recommend the following parameters:- temperature: 0\.2 - top\_k: 80 - repetition\_penalty: 1\.05

ModelDescriptionLFM2.5-8B-A1BOriginal model checkpoint in native format. Best for fine-tuning or inference with Transformers, vLLM, and SGLang.LFM2.5-8B-A1B-GGUFQuantized format for llama.cpp and compatible tools. Optimized for edge inference and local deployment.LFM2.5-8B-A1B-ONNXONNX Runtime format for cross-platform deployment.LFM2.5-8B-A1B-MLXMLX format for Apple Silicon. Optimized for fast inference on Mac devices. We recommend using LFM2.5-8B-A1B for agentic workflows, tool use, structured outputs, multilingual assistants, and on-device personal-assistant applications. It is not the best fit for heavy programming or knowledge-intensive question answering without retrieval.

https://huggingface.co/LiquidAI/LFM2.5-8B-A1B#chat-templateChat Template

LFM2.5 uses a ChatML-like format. See theChat Template documentationfor details. Example:

<|startoftext|><|im_start|>system
You are a helpful assistant trained by Liquid AI.<|im_end|>
<|im_start|>user
What is C. elegans?<|im_end|>
<|im_start|>assistant

Because LFM2.5-8B-A1B is a reasoning model, assistant turns contain an explicit chain of thought before the final answer. You can usetokenizer\.apply\_chat\_template\(\)to format your messages automatically.

https://huggingface.co/LiquidAI/LFM2.5-8B-A1B#tool-useTool Use

LFM2.5 supports function calling in four steps:

1. Function definition: Provide the list of tools as a JSON object in the system prompt, or usetokenizer\.apply\_chat\_template\(\)withtools=\.\.\..
2. Function call: By default, LFM2.5 writes Pythonic function calls (a Python list between<\|tool\_call\_start\|\>and<\|tool\_call\_end\|\>special tokens), as the assistant answer. You can override this behavior by asking the model to output JSON function calls in the system prompt.
3. Function execution: Execute the call and return the result with thetoolrole.
4. Final answer: LFM2.5 interprets the tool output and returns a plain-text answer addressing the original prompt.

See theTool Use documentationfor the full guide. Example:

<|startoftext|><|im_start|>system
List of tools: [{"name": "get_candidate_status", "description": "Retrieves the current status of a candidate in the recruitment process", "parameters": {"type": "object", "properties": {"candidate_id": {"type": "string", "description": "Unique identifier for the candidate"}}, "required": ["candidate_id"]}}]<|im_end|>
<|im_start|>user
What is the current status of candidate ID 12345?<|im_end|>
<|im_start|>assistant
<|tool_call_start|>[get_candidate_status(candidate_id="12345")]<|tool_call_end|>Checking the current status of candidate ID 12345.<|im_end|>
<|im_start|>tool
[{"candidate_id": "12345", "status": "Interview Scheduled", "position": "Clinical Research Associate", "date": "2023-11-20"}]<|im_end|>
<|im_start|>assistant
The candidate with ID 12345 is currently in the "Interview Scheduled" stage for the position of Clinical Research Associate, with an interview date set for 2023-11-20.<|im_end|>

https://huggingface.co/LiquidAI/LFM2.5-8B-A1B#%F0%9F%8F%83-inference🏃 Inference

LFM2.5-8B-A1B is supported by many inference frameworks. See theInference documentationfor the full list.

Quick start with Transformers (compatible withtransformers\>=5\.0\.0):

from transformers import AutoModelForCausalLM, AutoTokenizer, TextStreamer

model_id = "LiquidAI/LFM2.5-8B-A1B"
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    device_map="auto",
    dtype="bfloat16",
#   attn_implementation="flash_attention_2" <- uncomment on compatible GPU
)
tokenizer = AutoTokenizer.from_pretrained(model_id)
streamer = TextStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)

prompt = "What is C. elegans?"

input_ids = tokenizer.apply_chat_template(
    [{"role": "user", "content": prompt}],
    add_generation_prompt=True,
    return_tensors="pt",
    tokenize=True,
)["input_ids"].to(model.device)

output = model.generate(
    input_ids,
    do_sample=True,
    temperature=0.2,
    top_k=80,
    repetition_penalty=1.05,
    max_new_tokens=8192,
    streamer=streamer,
)

https://huggingface.co/LiquidAI/LFM2.5-8B-A1B#%F0%9F%94%A7-fine-tuning🔧 Fine-Tuning

We recommend fine-tuning LFM2.5 for your specific use case to achieve the best results.

https://huggingface.co/LiquidAI/LFM2.5-8B-A1B#%F0%9F%93%8A-performance📊 Performance

https://huggingface.co/LiquidAI/LFM2.5-8B-A1B#improvements-over-lfm2-8b-a1bImprovements over LFM2-8B-A1B

Thanks to reasoning, scaled-up pre-training, and large-scale RL, LFM2.5-8B-A1B improves over its predecessor across the board:

BenchmarkLFM2-8B-A1BLFM2.5-8B-A1BΔAA-Omniscience Index-78.42-24.70+53.62AA-Omniscience Accuracy7.338.67+1.34AA-Omniscience Non-Hallucination Rate7.4663.47+56.01IFEval79.4491.84+12.40IFBench26.0056.47+30.47Multi-IF58.5479.93+21.39MATH50074.8088.76+13.96AIME2520.0042.53+22.53BFCLv345.0764.36+19.29BFCLv425.5248.50+22.98Tau² Telecom13.6088.07+74.47Tau² Retail7.0239.82+32.80

https://huggingface.co/LiquidAI/LFM2.5-8B-A1B#knowledge-and-instruction-followingKnowledge and instruction following

ModelParametersAA-Omni. IndexAA-Omni. AccuracyAA-Omni. Non-Halluc.IFEvalIFBenchMulti-IFLFM2.5-8B-A1B8B/A1B-24.708.6763.4791.8456.4779.93Granite-4.0-H-Tiny7B/A1B-75.509.376.3882.2321.2859.00Qwen3.5-4B4B-51.5317.2016.9987.8050.3867.43Qwen3-30B-A3B-Thinking-250730.5B/3.3B-51.3118.8013.8790.8251.1179.04Gemma-4-E2B-IT5.1B-727.0015.0582.9333.5369.70Gemma-4-E4B-IT8B-50.678.1036.0687.7439.4877.58Gemma-4-26B-A4B-IT26B/4B-62.0714.3710.7591.4047.2582.06gpt-oss-20b21B/3.6B-49.1714.5724.5086.7358.6576.64

https://huggingface.co/LiquidAI/LFM2.5-8B-A1B#math-and-agentic-workflowsMath and agentic workflows

ModelParametersMATH500AIME25AIME26BFCLv3BFCLv4Tau² TelecomTau² RetailLFM2.5-8B-A1B8B/A1B88.7642.5350.0064.7949.7388.0739.82Granite-4.0-H-Tiny7B/A1B59.204.933.3356.8928.5216.6718.42Qwen3.5-4B4B80.7654.2858.3371.0654.0187.7271.93Qwen3-30B-A3B-Thinking-250730.5B/3.3B86.4871.6766.6773.3950.5321.9356.14Gemma-4-E2B-IT5.1B64.00263056.4431.9122.3718.95Gemma-4-E4B-IT8B65.0034.3340.6757.3133.9226.7542.11

https://huggingface.co/LiquidAI/LFM2.5-8B-A1B#cpu-inferenceCPU Inference

https://huggingface.co/LiquidAI/LFM2.5-8B-A1B#gpu-inferenceGPU Inference

LFM2.5-8B-A1B is the fastest model in its size class, reaching18.5K output tokens per second at high concurrency, over 1.6B tokens per day on a single H100.

https://huggingface.co/LiquidAI/LFM2.5-8B-A1B#%F0%9F%93%AC-contact📬 Contact

• Got questions or want to connect?Join our Discord community.
• If you are interested in custom solutions with edge deployment, please contactour sales team.

https://huggingface.co/LiquidAI/LFM2.5-8B-A1B#citationCitation

@article{liquidAI20268BA1B,
  author  = {Liquid AI},
  title   = {LFM2.5-8B-A1B: Personal Assistant On Your Laptop},
  journal = {Liquid AI Blog},
  year    = {2026},
  note    = {www.liquid.ai/blog/lfm2-5-8b-a1b},
}

@article{liquidai2025lfm2,
  title   = {LFM2 Technical Report},
  author  = {Liquid AI},
  journal = {arXiv preprint arXiv:2511.23404},
  year    = {2025}
}