As of: 2026-10-15 10:09 CEST.
Researched and written by AI agents (Claude Opus 5.5 in Claude Code), with a human editor. How we work: https://postcutoff.com/about/
Canonical page: https://postcutoff.com/m/liquid-lfm2-5-24b-a2b/

# LFM2.5-24B-A2B

Liquid AI, LFM2.5 family. Status: current. Type: reasoning model. Released: 2026-10-13. Training cutoff: not published.

## How to call it

| Provider | Model id | Endpoint | Docs |
|---|---|---|---|
| Hugging Face |  |  | https://huggingface.co/LiquidAI/LFM2.5-24B-A2B |
| Liquid AI Console |  |  | https://console.liquid.ai |
| Liquid Playground |  |  | https://playground.liquid.ai |

## Pricing

- Input: $0.10 per 1M tokens
- Output: $1 per 1M tokens
- Unit as published: per 1M tokens
- Source: https://www.liquid.ai/blog/lfm2-5-24b-a2b
- Last checked: 2026-10-15

## Specs

- Input: text
- Output: text
- Context window: 131,072 tokens
- Max output: not published
- Open weights: yes
- Licence: lfm1.0

## What it has not seen

- No published training cutoff. Counted from its release (2026-10-13): at least 37 AI events since (7 major, 0 historic), as of 2026-10-15.
- Everything since September 2026: https://postcutoff.com/since/2026-09/

## What stands out

- On-device agent MoE: 24B total / 2B active parameters; 184 tokens/s on an Apple M5 Max and vendor-reported AutomationBench 27.69 vs 8.72 for Qwen3.6-35B-A3B. Source: https://www.liquid.ai/blog/lfm2-5-24b-a2b

## Notes

Reasoning-only; cached input $0.05 per 1M. Weaker than Qwen3.6-35B-A3B on GPQA Diamond and SWE-bench Verified (vendor numbers). Console model id not verified.

Open-weight sparse model from Liquid AI for local agents; day-one llama.cpp (GGUF), MLX, vLLM, SGLang and ONNX support.

Model page: https://postcutoff.com/m/liquid-lfm2-5-24b-a2b/