gpt-oss-120b

OpenAI · 120.4B (active 5.1B) · MoE · ctx 128K · Apache 2.0 · HuggingFace

gpt-oss-120b (OpenAI) is a 120.4B parameters MoE (5.1B active per token) open-weight model tracked on AI Hashrate. At 8K context we estimate roughly 68.24 GB VRAM for Q4 (weights 66.22 GB + KV 1.02 GB + 1 GB overhead) and 242.82 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 ≈ 998.2 tok/s (estimated); MI300X 192GB ≈ 661.3 tok/s (estimated); H200 141GB SXM5 ≈ 598.9 tok/s (estimated); H100 80GB SXM5 ≈ 418.0 tok/s (estimated); Gaudi 3 128GB ≈ 408.8 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	998.2 est.	Yes	0.025
MI300X 192GB	192	Q4	661.3 est.	Yes	0.044
H200 141GB SXM5	141	Q4	598.9 est.	Yes	0.017
B200 192GB SXM	192	FP16	490.4 est.	No	0.012
H100 80GB SXM5	80	Q4	418.0 est.	Yes	0.014
Gaudi 3 128GB	128	Q4	408.8 est.	Yes	0.033
MI300X 192GB	192	FP16	324.9 est.	No	0.022
A100 80GB SXM4	80	Q4	254.4 est.	Yes	0.017
RTX 6000 Ada 48GB	48	Q4	169.3 est.	No	0.025
H200 141GB SXM5	141	FP16	158.7 est.	No	0.005
L40S 48GB	48	Q4	152.4 est.	No	0.019
RTX A6000 48GB	48	Q4	135.5 est.	No	0.03
Mac Studio M3 Ultra 96GB	96	Q4	102.2 est.	Yes	0.026
Gaudi 3 128GB	128	FP16	89.2 est.	No	0.007
Mac mini M4 Pro 64GB	64	Q4	85.6 est.	No	0.039
Mac Studio M2 Ultra 192GB	192	Q4	79.7 measured	Yes	0.011
MacBook Pro M4 Max 128GB	128	Q4	68.1 est.	Yes	0.015
MacBook Pro M3 Max 128GB	128	Q4	49.9 est.	Yes	0.012
Mac Studio M2 Ultra 192GB	192	FP16	49.0 est.	No	0.007
RTX 4090 24GB	24	Q4	44.5 est.	No	0.028
RTX 4090D 24GB (China)	24	Q4	44.5 est.	No	0.028
RTX 3090 Ti 24GB	24	Q4	44.5 est.	No	0.022
RX 7900 XTX 24GB	24	Q4	42.3 est.	No	0.042
RTX 3090 24GB	24	Q4	41.3 est.	No	0.028
H100 80GB SXM5	80	FP16	35.6 est.	No	0.001
RX 7900 XT 20GB	20	Q4	24.5 est.	No	0.027
A100 80GB SXM4	80	FP16	21.7 est.	No	0.001
RTX 5090 32GB	32	Q4	20.0 measured	No	0.01
MacBook Pro M4 Max 128GB	128	FP16	14.9 est.	No	0.003
Mac Studio M3 Ultra 96GB	96	FP16	12.6 est.	No	0.003
MacBook Pro M3 Max 128GB	128	FP16	10.9 est.	No	0.003
Mac mini M4 32GB	32	Q4	9.4 est.	No	0.009
Mac mini M4 Pro 64GB	64	FP16	1.9 est.	No	0.001

Fit checks

• RTX 5090 32GB — does not fit · ≈20.0 tok/s Q4 measured
• RTX 3090 Ti 24GB — does not fit · ≈44.5 tok/s Q4 est.
• RTX 4090 24GB — does not fit · ≈44.5 tok/s Q4 est.
• RTX 6000 Ada 48GB — does not fit · ≈169.3 tok/s Q4 est.
• RX 7900 XTX 24GB — does not fit · ≈42.3 tok/s Q4 est.
• RTX 3090 24GB — does not fit · ≈41.3 tok/s Q4 est.
• L40S 48GB — does not fit · ≈152.4 tok/s Q4 est.
• Mac Studio M3 Ultra 96GB — fits · ≈102.2 tok/s Q4 est.
• Mac Studio M2 Ultra 192GB — fits · ≈79.7 tok/s Q4 measured
• RX 7900 XT 20GB — does not fit · ≈24.5 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: gpt-oss

gpt-oss gpt-oss-20b

All gpt-oss models →