Mistral-Small-4-119B-2603

Mistral · 119.0B (active 6.5B) · MoE · ctx 256K · apache-2.0 · HuggingFace

Mistral-Small-4-119B-2603 (Mistral) is a 119.0B parameters MoE (6.5B active per token) open-weight model tracked on AI Hashrate. At 8K context we estimate roughly 67.75 GB VRAM for Q4 (weights 65.45 GB + KV 1.3 GB + 1 GB overhead) and 240.3 GB for FP16. About 10 GPUs/accelerators in our catalog fully fit this model at Q4 under that context assumption. Fastest Q4 configs: B200 192GB SXM ≈ 783.2 tok/s (estimated); MI300X 192GB ≈ 518.9 tok/s (estimated); H200 141GB SXM5 ≈ 469.9 tok/s (estimated); H100 80GB SXM5 ≈ 328.0 tok/s (estimated); Gaudi 3 128GB ≈ 320.7 tok/s (estimated). Use the table below for VRAM fit and tok/s per dollar (list MSRP). Estimates follow memory-bandwidth math; measured rows override when present. See methodology for details. Methodology.

33 hardware configs — sorted by tok/s. Default fit context 8K.

Hardware	VRAM	Quant	tok/s	Fits?	tok/s/$
B200 192GB SXM	192	Q4	783.2 est.	Yes	0.02
MI300X 192GB	192	Q4	518.9 est.	Yes	0.035
H200 141GB SXM5	141	Q4	469.9 est.	Yes	0.013
B200 192GB SXM	192	FP16	392.9 est.	No	0.01
H100 80GB SXM5	80	Q4	328.0 est.	Yes	0.011
Gaudi 3 128GB	128	Q4	320.7 est.	Yes	0.026
MI300X 192GB	192	FP16	260.3 est.	No	0.017
A100 80GB SXM4	80	Q4	199.6 est.	Yes	0.013
RTX 6000 Ada 48GB	48	Q4	134.8 est.	No	0.02
H200 141GB SXM5	141	FP16	127.1 est.	No	0.004
L40S 48GB	48	Q4	121.3 est.	No	0.015
RTX 5090 32GB	32	Q4	111.8 est.	No	0.056
RTX A6000 48GB	48	Q4	107.8 est.	No	0.024
Mac Studio M3 Ultra 96GB	96	Q4	80.2 est.	Yes	0.02
Mac Studio M2 Ultra 192GB	192	Q4	78.3 est.	Yes	0.011
Gaudi 3 128GB	128	FP16	71.5 est.	No	0.006
Mac mini M4 Pro 64GB	64	Q4	68.1 est.	No	0.031
MacBook Pro M4 Max 128GB	128	Q4	53.5 est.	Yes	0.012
Mac Studio M2 Ultra 192GB	192	FP16	39.3 est.	No	0.006
MacBook Pro M3 Max 128GB	128	Q4	39.2 est.	Yes	0.009
RTX 4090 24GB	24	Q4	35.4 est.	No	0.022
RTX 4090D 24GB (China)	24	Q4	35.4 est.	No	0.022
RTX 3090 Ti 24GB	24	Q4	35.4 est.	No	0.018
RX 7900 XTX 24GB	24	Q4	33.7 est.	No	0.034
RTX 3090 24GB	24	Q4	32.9 est.	No	0.022
H100 80GB SXM5	80	FP16	28.6 est.	No	0.001
RX 7900 XT 20GB	20	Q4	19.5 est.	No	0.022
A100 80GB SXM4	80	FP16	17.4 est.	No	0.001
MacBook Pro M4 Max 128GB	128	FP16	11.9 est.	No	0.003
Mac Studio M3 Ultra 96GB	96	FP16	10.1 est.	No	0.003
MacBook Pro M3 Max 128GB	128	FP16	8.7 est.	No	0.002
Mac mini M4 32GB	32	Q4	7.5 est.	No	0.008
Mac mini M4 Pro 64GB	64	FP16	1.5 est.	No	0.001

Fit checks

• RTX 5090 32GB — does not fit · ≈111.8 tok/s Q4 est.
• RTX 3090 Ti 24GB — does not fit · ≈35.4 tok/s Q4 est.
• RTX 4090 24GB — does not fit · ≈35.4 tok/s Q4 est.
• RTX 6000 Ada 48GB — does not fit · ≈134.8 tok/s Q4 est.
• RX 7900 XTX 24GB — does not fit · ≈33.7 tok/s Q4 est.
• RTX 3090 24GB — does not fit · ≈32.9 tok/s Q4 est.
• L40S 48GB — does not fit · ≈121.3 tok/s Q4 est.
• Mac Studio M3 Ultra 96GB — fits · ≈80.2 tok/s Q4 est.
• Mac Studio M2 Ultra 192GB — fits · ≈78.3 tok/s Q4 est.
• RX 7900 XT 20GB — does not fit · ≈19.5 tok/s Q4 est.

12 popular retail GPUs; every row in the table above also links its Fits? verdict to the full fit check.

Same family: Mistral

Mistral 7B Mistral-7B-Instruct-v0.3
Mistral Medium 3.5 Mistral-Medium-3.5-128B
Devstral 2 Devstral-Small-2-24B-Instruct-2512

All Mistral models →