Mistral-Medium-3.5-128B

Mistral · 128.0B · ctx 256K · Modified MIT · HuggingFace

Mistral-Medium-3.5-128B (Mistral) is a 128.0B parameters open-weight model tracked on AI Hashrate. At 8K context we estimate roughly 97.0 GB VRAM for Q4 (weights 70.4 GB + KV 25.6 GB + 1 GB overhead) and 282.6 GB for FP16. About 7 GPUs/accelerators in our catalog fully fit this model at Q4 under that context assumption. Fastest Q4 configs: B200 192GB SXM ≈ 39.8 tok/s (estimated); MI300X 192GB ≈ 26.3 tok/s (estimated); H200 141GB SXM5 ≈ 23.9 tok/s (estimated); Gaudi 3 128GB ≈ 16.3 tok/s (estimated); Mac Studio M2 Ultra 192GB ≈ 4.0 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.

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

Hardware	VRAM	Quant	tok/s	Fits?	tok/s/$
B200 192GB SXM	192	Q4	39.8 est.	Yes	0.001
H100 80GB SXM5	80	Q4	32.4 est.	No	0.001
MI300X 192GB	192	Q4	26.3 est.	Yes	0.002
H200 141GB SXM5	141	Q4	23.9 est.	Yes	0.001
A100 80GB SXM4	80	Q4	19.7 est.	No	0.001
Gaudi 3 128GB	128	Q4	16.3 est.	Yes	0.001
B200 192GB SXM	192	FP16	14.4 est.	No	0.0
Mac Studio M3 Ultra 96GB	96	Q4	11.4 est.	No	0.003
MI300X 192GB	192	FP16	9.6 est.	No	0.001
H200 141GB SXM5	141	FP16	4.7 est.	No	0.0
Mac Studio M2 Ultra 192GB	192	Q4	4.0 est.	Yes	0.001
RTX 6000 Ada 48GB	48	Q4	3.3 est.	No	0.0
L40S 48GB	48	Q4	3.0 est.	No	0.0
RTX 5090 32GB	32	Q4	2.8 est.	No	0.001
MacBook Pro M4 Max 128GB	128	Q4	2.7 est.	Yes	0.001
RTX A6000 48GB	48	Q4	2.7 est.	No	0.001
Gaudi 3 128GB	128	FP16	2.6 est.	No	0.0
MacBook Pro M3 Max 128GB	128	Q4	2.0 est.	Yes	0.0
Mac mini M4 Pro 64GB	64	Q4	1.7 est.	No	0.001
Mac Studio M2 Ultra 192GB	192	FP16	1.4 est.	No	0.0
H100 80GB SXM5	80	FP16	1.0 est.	No	0.0
A100 80GB SXM4	80	FP16	0.6 est.	No	0.0
MacBook Pro M4 Max 128GB	128	FP16	0.4 est.	No	0.0
Mac Studio M3 Ultra 96GB	96	FP16	0.4 est.	No	0.0
MacBook Pro M3 Max 128GB	128	FP16	0.3 est.	No	0.0
Mac mini M4 32GB	32	Q4	0.2 est.	No	0.0

Fit checks

• RTX 5090 32GB — does not fit · ≈2.8 tok/s Q4 est.
• RTX 6000 Ada 48GB — does not fit · ≈3.3 tok/s Q4 est.
• L40S 48GB — does not fit · ≈3.0 tok/s Q4 est.
• Mac Studio M3 Ultra 96GB — does not fit · ≈11.4 tok/s Q4 est.
• Mac Studio M2 Ultra 192GB — fits · ≈4.0 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 Small 4 Mistral-Small-4-119B-2603
Devstral 2 Devstral-Small-2-24B-Instruct-2512

All Mistral models →