Liquid AI has released LFM2.5-2.6B, an agentic model designed to run entirely on-device, enabling planning, tool calling, and multi-step task execution across phones, laptops, PCs, and robots. The model features 2.69B total parameters, a 131,072-token context window, and a 128,000-token vocabulary, pre-trained on approximately 34 trillion tokens. Two checkpoints are available: LFM2.5-2.6B-Base for fine-tuning and LFM2.5-2.6B, post-trained for agentic workloads. Since inference stays local, data never leaves the device, and marginal costs per run are near zero. Liquid AI reports tool-use and instruction-following scores competitive with models nearly four times its size, making it a strong option for 2026 edge AI deployments.
Is It Deployable?
Yes. Both checkpoints are publicly available on Hugging Face under the lfm1.0 license. Weights ship in native, GGUF, MLX, and ONNX formats, with day-one support in llama.cpp, vLLM, SGLang, and LM Studio.
- Which Companies: Solo developers and startups can pilot on existing hardware, with the model decoding at 220 tokens/s on an M5 Max in under 2.5 GB. Mid-market teams can self-host on a single NVIDIA H100 SXM5 GPU, serving roughly 1.3B tokens per day. Enterprises and OEMs can push the same weights to device fleets via GGUF and ONNX. Fine-tuning is available via LoRA with TRL and Unsloth.
- Which Industries: Liquid AI targets automotive, consumer electronics, industrial robotics, healthcare, financial services, e-commerce, and defense. Regulated and air-gapped environments benefit most, as no prompt reaches a third-party API.
- Applications: The model is recommended for agentic workloads—such as real-time customer support, on-device automation, and privacy-sensitive data processing—where latency, cost, and data security are critical. Its 128K context window supports long-horizon reasoning, and tool calling integrates smoothly with existing workflows.
Key Advantages for 2026: As edge computing matures, LFM2.5-2.6B offers a rare combination of compact size, high performance, and full local control, reducing reliance on cloud infrastructure and aligning with stricter data sovereignty regulations.
For more details, visit the Liquid AI blog or Hugging Face model repository.
via MarkTechPost