LFM2.5-2.6B (Reasoning)

by Liquid AI

LFM2.5-2.6B (Reasoning) is a reasoning model from Liquid AI, released on August 4, 2026. It ranks #129 of 220 models by Intelligence Score with a score of 18.9.

LFM2.5-2.6B is Liquid AI's open-weight 2.6B dense hybrid text→text model (LFM2 architecture) post-trained for on-device agentic workloads: planning, native tool calling, and multi-step tasks. ~2.69B parameters, trained on ~34T tokens, 128K context (131,072 tokens), 128K vocabulary. Always thinks before answering. Day-one support for Transformers, vLLM, SGLang, llama.cpp, MLX, and ONNX. Hugging Face: LiquidAI/LFM2.5-2.6B (+ Base). License: LFM Open License v1.0 (lfm1.0). Catalog benches include AIME25, LiveCodeBench v6, IFBench, Multi-IF, BFCL-v4, PinchBench, Claw-Eval, τ³-Banking 0.0567 (Liquid harness; scale differs from some AA τ³ rows), and aa-omniscience-index −29.50 (signed). No catalog id for ToolSandbox 77.83, BrowseComp+ (OpenClaw) 26.89, or IFStruct 85.49 — recorded here only. Sources: https://www.liquid.ai/blog/lfm2-5-2-6b · https://huggingface.co/LiquidAI/LFM2.5-2.6B. No official hosted provider row in this catalog (no sibling Liquid/OpenRouter provider pattern present); OpenRouter free route liquid/lfm-2.5-2.6b:free exists externally but was not added.

Key specs

Compare on the AI leaderboard · All AI models

More models from Liquid AI

• LFM2.5-VL-3B — Liquid AI