LLM Reference

LFM2.5 230M

Released
2026-06-25
Last refreshed
2026-06-29
Status
Researched today
ProprietaryCommercial use: conditionalAgentsJSON / Tool use

LFM2.5 230M is a released agents and json / tool use model; evaluate it while provider pricing coverage matures.

Use it for

  • Teams evaluating agents and json / tool use
  • Workloads that can use a 32k context window

Do not use it for

  • Cost-sensitive launches that need sourced token pricing
  • Vision or document-understanding workloads
  • Teams that need a tracked hosted API route today
Specifications
Family
LFM-2.5
Released
2026-06-25
Context
32k
Parameters
230M
Architecture
Decoder Only
Knowledge cutoff
2024
Specialization
general
Openness
Proprietary
License
ProprietaryCommercial use: conditional
Training
Pretrained
Created by

Adaptive and efficient AI models

Cambridge, MA, United States
Founded 2023
Website
Pricing

No tracked provider token pricing is available yet.

About

LFM2.5-230M is Liquid AI's smallest model in the LFM2.5 family, released June 25, 2026. With 230 million parameters and a 32K-token context window, it is designed for fast on-device tool use and large-scale data extraction — not general reasoning. The hybrid architecture uses 14 layers: 8 double-gated LIV convolutional blocks and 6 grouped-query attention (GQA) blocks, trained on 19 trillion tokens with supervised fine-tuning distilled from LFM2.5-350M followed by DPO and multi-domain reinforcement learning. Achieves 213 tokens/second on Samsung Galaxy S25 Ultra CPU and 42 tokens/second on Raspberry Pi 5. Competes with and often beats models more than twice its size on instruction following (IFEval: 71.71) and data extraction (CaseReportBench: 22.51). Supports tool calling via llama.cpp, MLX, vLLM, SGLang, and ONNX. Available in base and instruction-tuned variants on Hugging Face under the lfm1.0 license. Supports 10 languages including English, Chinese, Arabic, and Japanese.

LFM2.5 230M is a proprietary model in the LFM-2.5 family. The structured metadata tracks a 32k-token context window, function calling, and tool use. No headline benchmark score is tracked for LFM2.5 230M yet.

Top use-case fit: coding, agents, and build tasks

Agents

Included by capability and metadata signals in the decision map.

JSON / Tool use

Included by capability and metadata signals in the decision map.

Provider price ladder

No tracked provider token pricing is available for this model yet.

Capabilities

Function CallingTool Use

Benchmark peer barsfor Agents

No task-mapped benchmark peers are available for this model yet.

Migration checks

No linked migration route is available for this model yet.