LFM2.5-24B-A2B
AvailableLFM2.5-24B-A2B is Liquid AI's largest LFM2.5 model, released 2026-10-13: MoE with 24B total / ~2.45B active parameters (64 experts, top-4 routing, 40 layers), 131,072-token context (up from 32K in LFM2-24B-A2B), 128K vocabulary, 16 languages, text generation aimed at on-device and tool-using agents. Open weights (base + post-trained) on Hugging Face under the LFM Open License v1.0 (free until $10M annual revenue); day-one llama.cpp / MLX / vLLM / SGLang / ONNX support. Liquid Console API $0.10 / $1.00 per 1M input/output. Vendor benchmarks vs Qwen3.6-35B-A3B: AutomationBench 27.69 vs 8.72, ToolSandbox 80.71 vs 75.02, BFCL v4 65.01 vs 65.00; trails on SWE-bench Verified (51.0 vs 66.2) and GPQA Diamond (74.42 vs 83.69); IFEval 96.75.
Specifications
- License
- Open weights · LFM Open License v1.0
- Weights
- Downloadable
- Architecture
- Mixture-of-Experts
- Parameters
- 24B · 2.45B active
- Context window
- 131K tokens
- Max output
- —
- Knowledge cutoff
- —
- Price (in / out, $/M)
- $0.1 / $1
- Modalities
- TextCode
Benchmarks
No benchmark scores recorded yet. Spotted some? Submit a correction.
Vendor-reported figures are claims until independently verified. See methodology.