Skip to content

Ministral 3 14B VRAM requirements

How much VRAM Ministral 3 14B needs
4-bit weights: 7.9 GBWith 8K context: 10.0 GBWith 32K context: 13.7 GB8-bit, 8K context: 15.9 GB
Parameters
13.9B
Longest context
256K
Released
Dec 2025
License
Apache 2.0
Strongest at
Coding, Agents, Docs

How much VRAM Ministral 3 14B needs

Memory Ministral 3 14B needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
As released (FP8)14.6 GB16.8 GB20.5 GB35.5 GB
8-bit13.8 GB15.9 GB19.7 GB34.7 GB
6-bit10.7 GB12.8 GB16.5 GB31.5 GB
5-bit9.2 GB11.3 GB15.1 GB30.1 GB
4-bit7.9 GB10.0 GB13.7 GB28.7 GB
3-bit6.3 GB8.5 GB12.2 GB27.2 GB
2-bit4.9 GB7.0 GB10.7 GB25.7 GB
1.6-bit2.7 GB4.8 GB8.5 GB23.5 GB

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

GPUs and Macs that run Ministral 3 14B

The best way Ministral 3 14B runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: Ministral 3 14B on RTX 4060 report
RTX 5060Too bigPDF: Ministral 3 14B on RTX 5060 report
RTX 3070Too bigPDF: Ministral 3 14B on RTX 3070 report
RTX 3080 10 GBToo bigPDF: Ministral 3 14B on RTX 3080 10 GB report
RTX 3060 12 GB4-bit16K30 tok/s2.2 sPDF: Ministral 3 14B on RTX 3060 12 GB report
Arc B5804-bit16K27 tok/s1.5 sPDF: Ministral 3 14B on Arc B580 report
RTX 40704-bit16K42 tok/s1.0 sPDF: Ministral 3 14B on RTX 4070 report
RTX 4070 Super4-bit16K42 tok/s0.8 sPDF: Ministral 3 14B on RTX 4070 Super report
RTX 50704-bit16K55 tok/s0.9 sPDF: Ministral 3 14B on RTX 5070 report
RTX 4060 Ti 16 GB6-bit20K18 tok/s1.3 sPDF: Ministral 3 14B on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB6-bit20K28 tok/s1.2 sPDF: Ministral 3 14B on RTX 5060 Ti 16 GB report
Arc A770 16 GB6-bit20K24 tok/s1.4 sPDF: Ministral 3 14B on Arc A770 16 GB report
Radeon RX 9070 XT6-bit20K33 tok/s0.8 sPDF: Ministral 3 14B on Radeon RX 9070 XT report
RTX 4070 Ti Super6-bit20K41 tok/s0.6 sPDF: Ministral 3 14B on RTX 4070 Ti Super report
RTX 4080 Super6-bit20K45 tok/s0.5 sPDF: Ministral 3 14B on RTX 4080 Super report
RTX 5070 Ti6-bit20K54 tok/s0.6 sPDF: Ministral 3 14B on RTX 5070 Ti report
RTX 50806-bit20K57 tok/s0.5 sPDF: Ministral 3 14B on RTX 5080 report
Radeon RX 7900 XTX8-bit48K38 tok/s1.2 sPDF: Ministral 3 14B on Radeon RX 7900 XTX report
RTX 30908-bit48K44 tok/s0.8 sPDF: Ministral 3 14B on RTX 3090 report
RTX 40908-bit48K47 tok/s0.3 sPDF: Ministral 3 14B on RTX 4090 report
Radeon AI PRO R97008-bit96K26 tok/s0.8 sPDF: Ministral 3 14B on Radeon AI PRO R9700 report
RTX 50908-bit96K80 tok/s0.3 sPDF: Ministral 3 14B on RTX 5090 report
2× RTX 30908-bit192K44 tok/s0.5 sPDF: Ministral 3 14B on 2× RTX 3090 report
2× RTX 40908-bit192K47 tok/s0.2 sPDF: Ministral 3 14B on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit192K62 tok/s0.4 sPDF: Ministral 3 14B on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit192K45 tok/s0.3 sPDF: Ministral 3 14B on RTX 6000 Ada report
L40S8-bit192K41 tok/s0.5 sPDF: Ministral 3 14B on L40S report
2× RTX 50908-bit256K80 tok/s0.2 sPDF: Ministral 3 14B on 2× RTX 5090 report
4× RTX 30908-bit256K44 tok/s0.2 sPDF: Ministral 3 14B on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit256K80 tok/s0.2 sPDF: Ministral 3 14B on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB3-bit16K14 tok/s18 sPDF: Ministral 3 14B on Mac mini M4 16 GB report
Mac mini M4 24 GB6-bit24K8.8 tok/s18 sPDF: Ministral 3 14B on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit36K6.8 tok/s18 sPDF: Ministral 3 14B on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB6-bit24K19 tok/s8.9 sPDF: Ministral 3 14B on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit120K15 tok/s8.9 sPDF: Ministral 3 14B on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit192K15 tok/s8.9 sPDF: Ministral 3 14B on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit256K27 tok/s4.5 sPDF: Ministral 3 14B on M4 Max 128 GB report
M5 Max 128 GB8-bit256K30 tok/s1.3 sPDF: Ministral 3 14B on M5 Max 128 GB report
M3 Ultra 96 GB8-bit256K34 tok/s2.5 sPDF: Ministral 3 14B on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit256K34 tok/s2.5 sPDF: Ministral 3 14B on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit256K34 tok/s2.5 sPDF: Ministral 3 14B on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit256K47 tok/s0.8 sPDF: Ministral 3 14B on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit192K11 tok/s2.6 sPDF: Ministral 3 14B on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit256K11 tok/s2.6 sPDF: Ministral 3 14B on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit256K15 tok/s0.9 sPDF: Ministral 3 14B on RTX Spark report
DGX Spark8-bit256K13 tok/s0.9 sPDF: Ministral 3 14B on DGX Spark report
Dual DGX Spark8-bit256K13 tok/s0.5 sPDF: Ministral 3 14B on Dual DGX Spark report
DGX Station GB3008-bit256K248 tok/s0.1 sPDF: Ministral 3 14B on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit256K80 tok/s0.1 sPDF: Ministral 3 14B on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit256K80 tok/s0.1 sPDF: Ministral 3 14B on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit256K80 tok/s0.1 sPDF: Ministral 3 14B on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit256K90 tok/s0.6 sPDF: Ministral 3 14B on A100 80 GB SXM report
H100 SXM8-bit256K139 tok/s0.2 sPDF: Ministral 3 14B on H100 SXM report
4× H200 NVL8-bit256K185 tok/s0.1 sPDF: Ministral 3 14B on 4× H200 NVL report
8× H200 NVL8-bit256K185 tok/s0.1 sPDF: Ministral 3 14B 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 ministral-3:14b
llama.cpp
llama-server -hf unsloth/Ministral-3-14B-Instruct-2512-GGUF
vLLM
vllm serve mistralai/Ministral-3-14B-Instruct-2512
MLX (Mac)
mlx_lm.chat --model mlx-community/Ministral-3-14B-Instruct-2512-4bit

Questions

How much VRAM does Ministral 3 14B need?

Ministral 3 14B needs 7.9 GB for its weights at 4-bit, 10.0 GB with an 8K context and 13.7 GB with 32K. At 8-bit, 15.9 GB.

What is the minimum VRAM for Ministral 3 14B?

About 8.5 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 3060 12 GB.

How big is Ministral 3 14B?

13.9B parameters: 14.6 GB of weights as released (FP8), 7.9 GB at 4-bit.

Can Ministral 3 14B run on an RTX 4090?

Yes. At 8-bit it holds 48K of context and writes about 47 tokens a second.

More models