Skip to content

LFM2.5 24B A2B VRAM requirements

How much VRAM LFM2.5 24B A2B needs
4-bit weights: 13.5 GBWith 8K context: 14.4 GBWith 32K context: 14.9 GB8-bit, 8K context: 24.6 GB
Parameters
24B
Active per token
2B
Longest context
128K
Released
Oct 2026
License
LFM Open License v1.0
Strongest at
Agents

How much VRAM LFM2.5 24B A2B needs

Memory LFM2.5 24B A2B needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
16-bit44.6 GB45.6 GB46.0 GB47.9 GB
8-bit23.7 GB24.6 GB25.1 GB27.0 GB
6-bit18.3 GB19.2 GB19.7 GB21.6 GB
5-bit15.9 GB16.8 GB17.3 GB19.1 GB
4-bit13.5 GB14.4 GB14.9 GB16.8 GB
3-bit10.9 GB11.8 GB12.3 GB14.2 GB
2-bit8.4 GB9.3 GB9.8 GB11.6 GB
1.6-bit4.6 GB5.5 GB6.0 GB7.9 GB

One conversation, 16-bit context cache, runtime overhead included.

GPUs and Macs that run LFM2.5 24B A2B

The best way LFM2.5 24B A2B runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: LFM2.5 24B A2B on RTX 4060 report
RTX 5060Too bigPDF: LFM2.5 24B A2B on RTX 5060 report
RTX 3070Too bigPDF: LFM2.5 24B A2B on RTX 3070 report
RTX 3080 10 GBToo bigPDF: LFM2.5 24B A2B on RTX 3080 10 GB report
RTX 3060 12 GBToo bigPDF: LFM2.5 24B A2B on RTX 3060 12 GB report
Arc B580Too bigPDF: LFM2.5 24B A2B on Arc B580 report
RTX 4070Too bigPDF: LFM2.5 24B A2B on RTX 4070 report
RTX 4070 SuperToo bigPDF: LFM2.5 24B A2B on RTX 4070 Super report
RTX 5070Too bigPDF: LFM2.5 24B A2B on RTX 5070 report
RTX 4060 Ti 16 GB4-bit44K119 tok/s0.6 sPDF: LFM2.5 24B A2B on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB4-bit44K160 tok/s0.6 sPDF: LFM2.5 24B A2B on RTX 5060 Ti 16 GB report
Arc A770 16 GB4-bit44K120 tok/s0.7 sPDF: LFM2.5 24B A2B on Arc A770 16 GB report
Radeon RX 9070 XT4-bit44K148 tok/s0.4 sPDF: LFM2.5 24B A2B on Radeon RX 9070 XT report
RTX 4070 Ti Super4-bit44K201 tok/s0.3 sPDF: LFM2.5 24B A2B on RTX 4070 Ti Super report
RTX 4080 Super4-bit44K210 tok/s0.3 sPDF: LFM2.5 24B A2B on RTX 4080 Super report
RTX 5070 Ti4-bit44K231 tok/s0.3 sPDF: LFM2.5 24B A2B on RTX 5070 Ti report
RTX 50804-bit44K238 tok/s0.2 sPDF: LFM2.5 24B A2B on RTX 5080 report
Radeon RX 7900 XTX6-bit128K155 tok/s0.6 sPDF: LFM2.5 24B A2B on Radeon RX 7900 XTX report
RTX 30906-bit128K204 tok/s0.4 sPDF: LFM2.5 24B A2B on RTX 3090 report
RTX 40906-bit128K212 tok/s0.2 sPDF: LFM2.5 24B A2B on RTX 4090 report
Radeon AI PRO R97008-bit128K109 tok/s0.4 sPDF: LFM2.5 24B A2B on Radeon AI PRO R9700 report
RTX 50908-bit128K244 tok/s0.1 sPDF: LFM2.5 24B A2B on RTX 5090 report
2× RTX 30908-bit128K177 tok/s0.2 sPDF: LFM2.5 24B A2B on 2× RTX 3090 report
2× RTX 40908-bit128K185 tok/s0.1 sPDF: LFM2.5 24B A2B on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit128K215 tok/s0.2 sPDF: LFM2.5 24B A2B on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit128K180 tok/s0.2 sPDF: LFM2.5 24B A2B on RTX 6000 Ada report
L40S8-bit128K169 tok/s0.2 sPDF: LFM2.5 24B A2B on L40S report
2× RTX 50908-bit128K244 tok/s0.1 sPDF: LFM2.5 24B A2B on 2× RTX 5090 report
4× RTX 30908-bit128K177 tok/s0.1 sPDF: LFM2.5 24B A2B on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit128K244 tok/s0.1 sPDF: LFM2.5 24B A2B on RTX PRO 6000 Blackwell report
Mac mini M4 16 GBToo bigPDF: LFM2.5 24B A2B on Mac mini M4 16 GB report
Mac mini M4 24 GB4-bit56K56 tok/s8.7 sPDF: LFM2.5 24B A2B on Mac mini M4 24 GB report
Mac mini M4 32 GB6-bit64K45 tok/s8.7 sPDF: LFM2.5 24B A2B on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB4-bit56K91 tok/s4.4 sPDF: LFM2.5 24B A2B on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit128K66 tok/s4.4 sPDF: LFM2.5 24B A2B on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit128K66 tok/s4.4 sPDF: LFM2.5 24B A2B on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit128K93 tok/s2.2 sPDF: LFM2.5 24B A2B on M4 Max 128 GB report
M5 Max 128 GB8-bit128K98 tok/s0.6 sPDF: LFM2.5 24B A2B on M5 Max 128 GB report
M3 Ultra 96 GB8-bit128K104 tok/s1.2 sPDF: LFM2.5 24B A2B on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit128K104 tok/s1.2 sPDF: LFM2.5 24B A2B on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit128K104 tok/s1.2 sPDF: LFM2.5 24B A2B on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit128K120 tok/s0.4 sPDF: LFM2.5 24B A2B on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit128K57 tok/s1.3 sPDF: LFM2.5 24B A2B on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit128K57 tok/s1.3 sPDF: LFM2.5 24B A2B on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit128K80 tok/s0.5 sPDF: LFM2.5 24B A2B on RTX Spark report
DGX Spark8-bit128K74 tok/s0.5 sPDF: LFM2.5 24B A2B on DGX Spark report
Dual DGX Spark8-bit128K74 tok/s0.3 sPDF: LFM2.5 24B A2B on Dual DGX Spark report
DGX Station GB3008-bit128K354 tok/s0.1 sPDF: LFM2.5 24B A2B on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit128K244 tok/s0.1 sPDF: LFM2.5 24B A2B on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit128K244 tok/s0.1 sPDF: LFM2.5 24B A2B on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit128K244 tok/s0.1 sPDF: LFM2.5 24B A2B on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit128K257 tok/s0.3 sPDF: LFM2.5 24B A2B on A100 80 GB SXM report
H100 SXM8-bit128K302 tok/s0.1 sPDF: LFM2.5 24B A2B on H100 SXM report
4× H200 NVL8-bit128K330 tok/s0.1 sPDF: LFM2.5 24B A2B on 4× H200 NVL report
8× H200 NVL8-bit128K330 tok/s0.1 sPDF: LFM2.5 24B A2B on 8× H200 NVL report

Best fit is the highest quality, up to 8-bit, that leaves 16K of context. First reply is the wait for a 2K prompt. Speeds are rough, plus or minus 25%.

How to run it

Ollama
ollama run lfm2.5:24b
llama.cpp
llama-server -hf unsloth/LFM2.5-24B-A2B-GGUF
vLLM
vllm serve LiquidAI/LFM2.5-24B-A2B
MLX (Mac)
mlx_lm.chat --model mlx-community/LFM2.5-24B-A2B-OptiQ-4bit

Questions

How much VRAM does LFM2.5 24B A2B need?

LFM2.5 24B A2B needs 13.5 GB for its weights at 4-bit, 14.4 GB with an 8K context and 14.9 GB with 32K. At 8-bit, 24.6 GB.

What is the minimum VRAM for LFM2.5 24B A2B?

About 11.8 GB at 3-bit with an 8K context, the lowest precision we recommend for a model this size. The smallest machine here that runs it without spilling is the RTX 4060 Ti 16 GB.

How big is LFM2.5 24B A2B?

24B parameters (2B active per token): 44.6 GB of weights at 16-bit, 13.5 GB at 4-bit.

Can LFM2.5 24B A2B run on an RTX 4090?

Yes. At 6-bit it holds 128K of context and writes about 212 tokens a second.

More models