Skip to content

Mistral Small 4 119B VRAM requirements

How much VRAM Mistral Small 4 119B needs
4-bit weights: 67.4 GBWith 8K context: 68.4 GBWith 32K context: 68.9 GB
Parameters
119B
Active per token
6.5B
Longest context
256K
Released
Mar 2026
License
Apache 2.0
Strongest at
Chat, Docs

How much VRAM Mistral Small 4 119B needs

Memory Mistral Small 4 119B needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
As released (FP8)113 GB114 GB114 GB116 GB
6-bit91.2 GB92.2 GB92.7 GB94.8 GB
5-bit79.1 GB80.1 GB80.6 GB82.7 GB
4-bit67.4 GB68.4 GB68.9 GB71.1 GB
3-bit54.3 GB55.3 GB55.9 GB58.0 GB
2-bit41.7 GB42.7 GB43.2 GB45.3 GB
1.6-bit22.9 GB23.9 GB24.5 GB26.6 GB

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

GPUs and Macs that run Mistral Small 4 119B

The best way Mistral Small 4 119B runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: Mistral Small 4 119B on RTX 4060 report
RTX 5060Too bigPDF: Mistral Small 4 119B on RTX 5060 report
RTX 3070Too bigPDF: Mistral Small 4 119B on RTX 3070 report
RTX 3080 10 GBToo bigPDF: Mistral Small 4 119B on RTX 3080 10 GB report
RTX 3060 12 GBToo bigPDF: Mistral Small 4 119B on RTX 3060 12 GB report
Arc B580Too bigPDF: Mistral Small 4 119B on Arc B580 report
RTX 4070Too bigPDF: Mistral Small 4 119B on RTX 4070 report
RTX 4070 SuperToo bigPDF: Mistral Small 4 119B on RTX 4070 Super report
RTX 5070Too bigPDF: Mistral Small 4 119B on RTX 5070 report
RTX 4060 Ti 16 GBToo bigPDF: Mistral Small 4 119B on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GBToo bigPDF: Mistral Small 4 119B on RTX 5060 Ti 16 GB report
Arc A770 16 GBToo bigPDF: Mistral Small 4 119B on Arc A770 16 GB report
Radeon RX 9070 XTToo bigPDF: Mistral Small 4 119B on Radeon RX 9070 XT report
RTX 4070 Ti SuperToo bigPDF: Mistral Small 4 119B on RTX 4070 Ti Super report
RTX 4080 SuperToo bigPDF: Mistral Small 4 119B on RTX 4080 Super report
RTX 5070 TiToo bigPDF: Mistral Small 4 119B on RTX 5070 Ti report
RTX 5080Too bigPDF: Mistral Small 4 119B on RTX 5080 report
Radeon RX 7900 XTXToo bigPDF: Mistral Small 4 119B on Radeon RX 7900 XTX report
RTX 3090Too bigPDF: Mistral Small 4 119B on RTX 3090 report
RTX 4090Too bigPDF: Mistral Small 4 119B on RTX 4090 report
Radeon AI PRO R9700Too bigPDF: Mistral Small 4 119B on Radeon AI PRO R9700 report
RTX 5090Too bigPDF: Mistral Small 4 119B on RTX 5090 report
2× RTX 30902-bit104K170 tok/s0.7 sPDF: Mistral Small 4 119B on 2× RTX 3090 report
2× RTX 40902-bit104K178 tok/s0.3 sPDF: Mistral Small 4 119B on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB2-bit136K210 tok/s0.6 sPDF: Mistral Small 4 119B on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada2-bit136K173 tok/s0.4 sPDF: Mistral Small 4 119B on RTX 6000 Ada report
L40S2-bit136K162 tok/s0.7 sPDF: Mistral Small 4 119B on L40S report
2× RTX 50903-bit224K213 tok/s0.2 sPDF: Mistral Small 4 119B on 2× RTX 5090 report
4× RTX 30905-bit256K109 tok/s0.4 sPDF: Mistral Small 4 119B on 4× RTX 3090 report
RTX PRO 6000 Blackwell5-bit256K171 tok/s0.3 sPDF: Mistral Small 4 119B on RTX PRO 6000 Blackwell report
Mac mini M4 16 GBToo bigPDF: Mistral Small 4 119B on Mac mini M4 16 GB report
Mac mini M4 24 GBToo bigPDF: Mistral Small 4 119B on Mac mini M4 24 GB report
Mac mini M4 32 GBToo bigPDF: Mistral Small 4 119B on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GBToo bigPDF: Mistral Small 4 119B on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GBToo bigPDF: Mistral Small 4 119B on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB2-bit144K63 tok/s13 sPDF: Mistral Small 4 119B on Mac mini M4 Pro 64 GB report
M4 Max 128 GB5-bit256K61 tok/s6.5 sPDF: Mistral Small 4 119B on M4 Max 128 GB report
M5 Max 128 GB5-bit256K66 tok/s1.9 sPDF: Mistral Small 4 119B on M5 Max 128 GB report
M3 Ultra 96 GB4-bit16K80 tok/s3.7 sPDF: Mistral Small 4 119B on M3 Ultra 96 GB report
M3 Ultra 256 GBFP8256K57 tok/s3.7 sPDF: Mistral Small 4 119B on M3 Ultra 256 GB report
M3 Ultra 512 GBFP8256K57 tok/s3.7 sPDF: Mistral Small 4 119B on M3 Ultra 512 GB report
M5 Ultra 512 GBFP8256K74 tok/s1.1 sPDF: Mistral Small 4 119B on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB2-bit144K52 tok/s3.8 sPDF: Mistral Small 4 119B on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB5-bit256K30 tok/s3.8 sPDF: Mistral Small 4 119B on Ryzen AI Max+ 395 128 GB report
RTX SparkFP848K30 tok/s2.3 sPDF: Mistral Small 4 119B on RTX Spark report
DGX SparkFP848K27 tok/s2.5 sPDF: Mistral Small 4 119B on DGX Spark report
Dual DGX SparkFP8256K27 tok/s1.4 sPDF: Mistral Small 4 119B on Dual DGX Spark report
DGX Station GB300FP8256K287 tok/s0.1 sPDF: Mistral Small 4 119B on DGX Station GB300 report
2× RTX PRO 6000 BlackwellFP8256K135 tok/s0.2 sPDF: Mistral Small 4 119B on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 BlackwellFP8256K135 tok/s0.1 sPDF: Mistral Small 4 119B on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 BlackwellFP8256K135 tok/s0.1 sPDF: Mistral Small 4 119B on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM4-bit256K203 tok/s0.8 sPDF: Mistral Small 4 119B on A100 80 GB SXM report
H100 SXM4-bit256K260 tok/s0.3 sPDF: Mistral Small 4 119B on H100 SXM report
4× H200 NVLFP8256K243 tok/s0.1 sPDF: Mistral Small 4 119B on 4× H200 NVL report
8× H200 NVLFP8256K243 tok/s0.1 sPDF: Mistral Small 4 119B 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

llama.cpp
llama-server -hf unsloth/Mistral-Small-4-119B-2603-GGUF
vLLM
vllm serve mistralai/Mistral-Small-4-119B-2603
MLX (Mac)
mlx_lm.chat --model mlx-community/Mistral-Small-4-119B-2603-4bit

Questions

How much VRAM does Mistral Small 4 119B need?

Mistral Small 4 119B needs 67.4 GB for its weights at 4-bit, 68.4 GB with an 8K context and 68.9 GB with 32K.

What is the minimum VRAM for Mistral Small 4 119B?

About 42.7 GB at 2-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 2× RTX 3090.

How big is Mistral Small 4 119B?

119B parameters (6.5B active per token): 113 GB of weights as released (FP8), 67.4 GB at 4-bit.

Can Mistral Small 4 119B run on an RTX 4090?

Not on one card. The smallest machine here that runs it is the 2× RTX 3090.

More models