ErtasAI/LFM2.5-1.2B-Instruct

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:1.2BQuant:BF16Context Size:32kPublished:Jul 16, 2026License:otherArchitecture:Transformer0.0K Featherless Exclusive Cold

LFM2.5-1.2B-Instruct is a 1.17 billion parameter instruction-tuned causal language model developed by Liquid AI, designed for efficient on-device deployment. This hybrid model, built on the LFM2 architecture, features extended pre-training on 28T tokens and large-scale multi-stage reinforcement learning, supporting a 32,768-token context length. It delivers best-in-class performance for its size, rivaling larger models, and offers fast edge inference on CPUs and mobile NPUs, making it suitable for agentic tasks, data extraction, and RAG in resource-constrained environments. The model supports multiple languages including English, Arabic, Chinese, French, German, Japanese, Korean, and Spanish, and includes advanced features like tool use and a ChatML-like chat template.

Loading preview...

LFM2.5-1.2B-Instruct: On-Device AI

LFM2.5-1.2B-Instruct, developed by Liquid AI, is a 1.17 billion parameter instruction-tuned model optimized for on-device deployment. It is part of the LFM2.5 family of hybrid models, building on the LFM2 architecture with significantly extended pre-training (28T tokens) and multi-stage reinforcement learning.

Key Capabilities & Features

  • Best-in-Class Performance: Achieves performance comparable to much larger models, enabling high-quality AI in resource-constrained environments.
  • Fast Edge Inference: Designed for rapid execution, demonstrating 239 tok/s decode on AMD CPU and 82 tok/s on mobile NPU, while operating under 1GB of memory.
  • Broad Compatibility: Offers day-one support for popular inference frameworks like llama.cpp, MLX, and vLLM, with quantized formats (GGUF, ONNX, MLX) for diverse hardware.
  • Multilingual Support: Capable in English, Arabic, Chinese, French, German, Japanese, Korean, and Spanish.
  • Advanced Functionality: Supports a ChatML-like chat template and robust tool use capabilities, including Pythonic function calls and JSON output options.
  • Extensive Context: Features a 32,768-token context length and a 65,536-token vocabulary size.

Performance Highlights

Benchmarks show LFM2.5-1.2B-Instruct outperforming other sub-2B models like Qwen3-1.7B and Gemma 3 1B IT across various metrics, including GPQA, MMLU-Pro, IFEval, and BFCLv3. Its efficient design also translates to superior inference speeds and lower memory consumption compared to similar-sized models on mobile CPUs and NPUs.

Recommended Use Cases

  • Agentic tasks
  • Data extraction
  • Retrieval-Augmented Generation (RAG)

It is not recommended for knowledge-intensive tasks or programming.