Skip to content

gpt-oss 120B VRAM requirements

How much VRAM gpt-oss 120B needs
MXFP4 weights: 59.0 GBWith 8K context: 60.2 GBWith 32K context: 61.1 GB
Parameters
117B
Active per token
5.1B
Longest context
128K
Released
Aug 2025
License
Apache 2.0
Strongest at
Chat, Coding

How much VRAM gpt-oss 120B needs

Memory gpt-oss 120B needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
As released (MXFP4)59.0 GB60.2 GB61.1 GB64.5 GB
3-bit53.2 GB54.4 GB55.2 GB58.6 GB
2-bit40.8 GB42.0 GB42.9 GB46.2 GB
1.6-bit22.4 GB23.6 GB24.5 GB27.9 GB

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

GPUs and Macs that run gpt-oss 120B

The best way gpt-oss 120B runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: gpt-oss 120B on RTX 4060 report
RTX 5060Too bigPDF: gpt-oss 120B on RTX 5060 report
RTX 3070Too bigPDF: gpt-oss 120B on RTX 3070 report
RTX 3080 10 GBToo bigPDF: gpt-oss 120B on RTX 3080 10 GB report
RTX 3060 12 GBToo bigPDF: gpt-oss 120B on RTX 3060 12 GB report
Arc B580Too bigPDF: gpt-oss 120B on Arc B580 report
RTX 4070Too bigPDF: gpt-oss 120B on RTX 4070 report
RTX 4070 SuperToo bigPDF: gpt-oss 120B on RTX 4070 Super report
RTX 5070Too bigPDF: gpt-oss 120B on RTX 5070 report
RTX 4060 Ti 16 GBToo bigPDF: gpt-oss 120B on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GBToo bigPDF: gpt-oss 120B on RTX 5060 Ti 16 GB report
Arc A770 16 GBToo bigPDF: gpt-oss 120B on Arc A770 16 GB report
Radeon RX 9070 XTToo bigPDF: gpt-oss 120B on Radeon RX 9070 XT report
RTX 4070 Ti SuperToo bigPDF: gpt-oss 120B on RTX 4070 Ti Super report
RTX 4080 SuperToo bigPDF: gpt-oss 120B on RTX 4080 Super report
RTX 5070 TiToo bigPDF: gpt-oss 120B on RTX 5070 Ti report
RTX 5080Too bigPDF: gpt-oss 120B on RTX 5080 report
Radeon RX 7900 XTXToo bigPDF: gpt-oss 120B on Radeon RX 7900 XTX report
RTX 3090Too bigPDF: gpt-oss 120B on RTX 3090 report
RTX 4090Too bigPDF: gpt-oss 120B on RTX 4090 report
Radeon AI PRO R9700Too bigPDF: gpt-oss 120B on Radeon AI PRO R9700 report
RTX 5090Too bigPDF: gpt-oss 120B on RTX 5090 report
2× RTX 30902-bit88K197 tok/s0.5 sPDF: gpt-oss 120B on 2× RTX 3090 report
2× RTX 40902-bit88K205 tok/s0.2 sPDF: gpt-oss 120B on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB2-bit104K239 tok/s0.4 sPDF: gpt-oss 120B on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada2-bit104K200 tok/s0.3 sPDF: gpt-oss 120B on RTX 6000 Ada report
L40S2-bit104K188 tok/s0.5 sPDF: gpt-oss 120B on L40S report
2× RTX 50903-bit128K241 tok/s0.2 sPDF: gpt-oss 120B on 2× RTX 5090 report
4× RTX 3090MXFP4128K129 tok/s0.3 sPDF: gpt-oss 120B on 4× RTX 3090 report
RTX PRO 6000 BlackwellMXFP4128K197 tok/s0.2 sPDF: gpt-oss 120B on RTX PRO 6000 Blackwell report
Mac mini M4 16 GBToo bigPDF: gpt-oss 120B on Mac mini M4 16 GB report
Mac mini M4 24 GBToo bigPDF: gpt-oss 120B on Mac mini M4 24 GB report
Mac mini M4 32 GBToo bigPDF: gpt-oss 120B on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GBToo bigPDF: gpt-oss 120B on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GBToo bigPDF: gpt-oss 120B on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB2-bit112K74 tok/s9.3 sPDF: gpt-oss 120B on Mac mini M4 Pro 64 GB report
M4 Max 128 GBMXFP4128K72 tok/s4.6 sPDF: gpt-oss 120B on M4 Max 128 GB report
M5 Max 128 GBMXFP4128K77 tok/s1.3 sPDF: gpt-oss 120B on M5 Max 128 GB report
M3 Ultra 96 GBMXFP4128K84 tok/s2.6 sPDF: gpt-oss 120B on M3 Ultra 96 GB report
M3 Ultra 256 GBMXFP4128K84 tok/s2.6 sPDF: gpt-oss 120B on M3 Ultra 256 GB report
M3 Ultra 512 GBMXFP4128K84 tok/s2.6 sPDF: gpt-oss 120B on M3 Ultra 512 GB report
M5 Ultra 512 GBMXFP4128K103 tok/s0.8 sPDF: gpt-oss 120B on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB2-bit112K64 tok/s2.7 sPDF: gpt-oss 120B on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GBMXFP4128K37 tok/s2.7 sPDF: gpt-oss 120B on Ryzen AI Max+ 395 128 GB report
RTX SparkMXFP4128K51 tok/s1.2 sPDF: gpt-oss 120B on RTX Spark report
DGX SparkMXFP4128K47 tok/s1.3 sPDF: gpt-oss 120B on DGX Spark report
Dual DGX SparkMXFP4128K47 tok/s0.8 sPDF: gpt-oss 120B on Dual DGX Spark report
DGX Station GB300MXFP4128K346 tok/s0.1 sPDF: gpt-oss 120B on DGX Station GB300 report
2× RTX PRO 6000 BlackwellMXFP4128K197 tok/s0.1 sPDF: gpt-oss 120B on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 BlackwellMXFP4128K197 tok/s0.1 sPDF: gpt-oss 120B on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 BlackwellMXFP4128K197 tok/s0.1 sPDF: gpt-oss 120B on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXMMXFP4128K212 tok/s0.6 sPDF: gpt-oss 120B on A100 80 GB SXM report
H100 SXMMXFP4128K269 tok/s0.2 sPDF: gpt-oss 120B on H100 SXM report
4× H200 NVLMXFP4128K308 tok/s0.1 sPDF: gpt-oss 120B on 4× H200 NVL report
8× H200 NVLMXFP4128K308 tok/s0.1 sPDF: gpt-oss 120B 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 gpt-oss:120b
llama.cpp
llama-server -hf unsloth/gpt-oss-120b-GGUF
vLLM
vllm serve openai/gpt-oss-120b
MLX (Mac)
mlx_lm.chat --model mlx-community/gpt-oss-120b-MXFP4-Q4

Questions

How much VRAM does gpt-oss 120B need?

gpt-oss 120B needs 59.0 GB for its weights as released (MXFP4), 60.2 GB with an 8K context and 61.1 GB with 32K.

What is the minimum VRAM for gpt-oss 120B?

About 42.0 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 gpt-oss 120B?

117B parameters (5.1B active per token): 59.0 GB of weights as released (MXFP4).

Can gpt-oss 120B run on an RTX 4090?

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

More models