LFM2.5 1.2B is Liquid AI's on-device model family built on hybrid LIV convolution and GQA, trained on 28T tokens for edge agentic AI under 1GB memory.
Capabilities, design details, and architectural traits
LFM2.5 1.2B is documented by Liquid AI as a model family built on the LFM2 device-optimized architecture, designed to be the building block of on-device agentic AI. It is not a single model but a full family, including Base, Instruct, Japanese, Vision-Language, and Audio-Language variants built on a shared backbone.
| Trait | Detail |
|---|---|
| Hybrid architecture | Built on a 16-layer hybrid combining 10 double-gated LIV convolution blocks with 6 Grouped Query Attention (GQA) blocks, designed for fast, memory-efficient inference on CPUs and NPUs rather than GPU clusters. |
| Scaled pretraining | Extended from 10T to 28T tokens compared to LFM2, combined with multi-stage reinforcement learning focused on instruction following, tool use, math, and knowledge reasoning. |
| Sub-1GB memory footprint | Documented as running under 1GB of memory in quantized form with day-one support for llama.cpp, MLX, vLLM, and NPU targets including AMD, Qualcomm, and Apple Silicon. |
| Complete specialist family | Alongside the general Instruct variant, ships with a Japanese-optimized chat model, a VL model with multilingual vision, and an Audio model documented as 8x faster than its predecessor. |
Liquid AI documents LFM2.5 as free to download, run, and fine-tune commercially until a company exceeds $10M in annual revenue, a usage condition distinct from a standard open-source license.
Independent evaluations · Artificial Analysis
Evaluate specifications, pricing, and independent benchmark indices
| Model Details | |||
|---|---|---|---|
| General Info | |||
| Provider | Liquid AI | Anthropic | Anthropic |
| Release Date | January 5, 2026 | July 24, 2026 | June 9, 2026 |
| Knowledge Cutoff | - | May 2026 | - |
| Context & Limits | |||
| Context Window | 33K | 1M Best Context Window | 1M Best Context Window |
| Pricing (per 1M tokens) | |||
| Input Pricing | Free Best Input Pricing | $5 | $10 |
| Output Pricing | Free Best Output Pricing | $25 | $50 |
| Modalities | |||
| Inputs | text | textimage | textimagefile |
| Outputs | text | text | text |
| Benchmarks (0-100) | |||
| Intelligence Index | 2.3 | 63.1 Best Intelligence Index | 62.1 |
| Coding Index | - | 78.0 Best Coding Index | 76.5 |
| Agentic Index | - | 59.2 Best Agentic Index | 56.6 |
Graduate-level reasoning and expert Q&A evaluation.
Extremely difficult logical reasoning and knowledge.
Logical reasoning over long context windows.
Independent evaluation data provided by Artificial Analysis. To view the latest benchmarks and full details, visit their official site.
