Liquid AI releases open-weight LFM2.5-24B-A2B agent model

Liquid AI unveiled LFM2.5-24B-A2B, an open-weight Mixture-of-Experts model that stores 24B parameters but activates about 2B per token. It raises context from 32K to 131,072 tokens, doubles the vocabulary to 128K across 16 languages, and ships with day-one support for llama.cpp, MLX, vLLM, SGLang and ONNX.

• 24B parameters with ~2B activated per token via MoE
• Context window increased from 32K to 131,072 tokens
• Vocabulary doubled to 128K tokens, 16 languages supported
• Leads Liquid AI's own benchmarks on tool use and instruction following