LFM2.5 24B A2B VRAM requirements
Liquid AI · 24B params (2.45B active) · LFM Open License v1.0. Estimates, not vendor numbers.
LFM2.5 24B A2B (24B parameters, 2.45B active per token) needs about 16 GB of VRAM or unified memory at Q4, 19 GB at Q5 and 27 GB at Q8 (weights plus about 1.5 GB headroom, our estimate). The smallest common setup that runs it is 24GB GPU (RTX 4090).
| Quant | Download | Memory needed |
|---|---|---|
| Q4 | 14 GB | 16 GB |
| Q5 | 17 GB | 19 GB |
| Q8 | 26 GB | 27 GB |
Will it run on my machine?
| Setup | Runs? | Speed est. |
|---|---|---|
| 8GB GPU (RTX 4060) | No | n/a |
| 12GB GPU (RTX 3060 / 4070) | No | n/a |
| 16GB GPU (RTX 4060 Ti 16GB) | No | n/a |
| 24GB GPU (RTX 4090) | Yes, Q5 | ~346 tok/s |
| 32GB GPU (RTX 5090) | Yes, Q8 | ~413 tok/s |
| Mac 16GB (M4) | No | n/a |
| Mac 48GB (M4 Pro) | Yes, Q8 | ~63 tok/s |
| Mac 128GB (M4 Max) | Yes, Q8 | ~126 tok/s |
| PC, 32GB RAM, no GPU | Yes, Q5 | ~21 tok/s |
Best quant that fits each setup, with estimated generation speed from memory bandwidth. Real speed depends on the runtime, context length and settings.
Mixture of experts: 24B total, 2.45B active per token, 128K context. Reasoning model built for on-device agents and tool calling; Liquid AI says pair it with search for knowledge-heavy questions.
FAQ
How much VRAM does LFM2.5 24B A2B need?
About 16 GB at Q4, 19 GB at Q5 and 27 GB at Q8, including about 1.5 GB headroom. These are estimates from the parameter count on the official model card.
Can LFM2.5 24B A2B run on 8GB of VRAM?
Not fully on the GPU. At Q4 it needs about 16 GB; partial CPU offload works but is much slower.
Can LFM2.5 24B A2B run on a Mac?
Yes. Mac 48GB (M4 Pro), Mac 128GB (M4 Max) can run it.
Which quant of LFM2.5 24B A2B should I use?
Use the highest quant that fits with room to spare: Q8 is near full quality, Q5 is a good balance, and Q4 is the smallest that usually keeps quality acceptable.
Why is LFM2.5 24B A2B fast for its size?
It is a mixture of experts: 24B parameters in total but only 2.45B are used per token, so it reads less memory per token. You still need memory for all 24B.
Similar-size models
gpt-oss-20b (21B) · Gemma 4 26B A4B (25.2B) · Qwen3.6 27B (27B) · Qwen3.8 27B (27.8B) · Muse Glimmer 30B (29.6B) · Qwen3 Coder 30B A3B (30.5B)