Skip to content

gpt-oss 20B VRAM requirements

How much VRAM gpt-oss 20B needs
4-bit weights: 11.8 GBWith 8K context: 12.9 GBWith 32K context: 13.5 GB
Parameters
20.9B
Active per token
3.6B
Longest context
128K
Released
Aug 2025
License
Apache 2.0
Strongest at
Agents

How much VRAM gpt-oss 20B needs

Memory gpt-oss 20B needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
As released (MXFP4)11.3 GB12.4 GB13.0 GB15.2 GB
4-bit11.8 GB12.9 GB13.5 GB15.7 GB
3-bit9.5 GB10.6 GB11.2 GB13.4 GB
2-bit7.3 GB8.4 GB9.0 GB11.2 GB
1.6-bit4.0 GB5.1 GB5.7 GB7.9 GB

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

GPUs and Macs that run gpt-oss 20B

The best way gpt-oss 20B runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: gpt-oss 20B on RTX 4060 report
RTX 5060Too bigPDF: gpt-oss 20B on RTX 5060 report
RTX 3070Too bigPDF: gpt-oss 20B on RTX 3070 report
RTX 3080 10 GBToo bigPDF: gpt-oss 20B on RTX 3080 10 GB report
RTX 3060 12 GB3-bit40K119 tok/s1.4 sPDF: gpt-oss 20B on RTX 3060 12 GB report
Arc B5803-bit40K99 tok/s1.0 sPDF: gpt-oss 20B on Arc B580 report
RTX 40703-bit40K156 tok/s0.6 sPDF: gpt-oss 20B on RTX 4070 report
RTX 4070 Super3-bit40K156 tok/s0.5 sPDF: gpt-oss 20B on RTX 4070 Super report
RTX 50703-bit40K193 tok/s0.6 sPDF: gpt-oss 20B on RTX 5070 report
RTX 4060 Ti 16 GBMXFP4120K71 tok/s0.8 sPDF: gpt-oss 20B on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GBMXFP4120K104 tok/s0.8 sPDF: gpt-oss 20B on RTX 5060 Ti 16 GB report
Arc A770 16 GBMXFP4120K86 tok/s0.9 sPDF: gpt-oss 20B on Arc A770 16 GB report
Radeon RX 9070 XTMXFP4120K113 tok/s0.5 sPDF: gpt-oss 20B on Radeon RX 9070 XT report
RTX 4070 Ti SuperMXFP4120K145 tok/s0.4 sPDF: gpt-oss 20B on RTX 4070 Ti Super report
RTX 4080 SuperMXFP4120K156 tok/s0.3 sPDF: gpt-oss 20B on RTX 4080 Super report
RTX 5070 TiMXFP4120K181 tok/s0.4 sPDF: gpt-oss 20B on RTX 5070 Ti report
RTX 5080MXFP4120K190 tok/s0.3 sPDF: gpt-oss 20B on RTX 5080 report
Radeon RX 7900 XTXMXFP4128K151 tok/s0.8 sPDF: gpt-oss 20B on Radeon RX 7900 XTX report
RTX 3090MXFP4128K187 tok/s0.5 sPDF: gpt-oss 20B on RTX 3090 report
RTX 4090MXFP4128K197 tok/s0.2 sPDF: gpt-oss 20B on RTX 4090 report
Radeon AI PRO R9700MXFP4128K113 tok/s0.5 sPDF: gpt-oss 20B on Radeon AI PRO R9700 report
RTX 5090MXFP4128K287 tok/s0.2 sPDF: gpt-oss 20B on RTX 5090 report
2× RTX 3090MXFP4128K187 tok/s0.3 sPDF: gpt-oss 20B on 2× RTX 3090 report
2× RTX 4090MXFP4128K197 tok/s0.1 sPDF: gpt-oss 20B on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GBMXFP4128K240 tok/s0.3 sPDF: gpt-oss 20B on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 AdaMXFP4128K190 tok/s0.2 sPDF: gpt-oss 20B on RTX 6000 Ada report
L40SMXFP4128K176 tok/s0.3 sPDF: gpt-oss 20B on L40S report
2× RTX 5090MXFP4128K287 tok/s0.1 sPDF: gpt-oss 20B on 2× RTX 5090 report
4× RTX 3090MXFP4128K187 tok/s0.2 sPDF: gpt-oss 20B on 4× RTX 3090 report
RTX PRO 6000 BlackwellMXFP4128K287 tok/s0.1 sPDF: gpt-oss 20B on RTX PRO 6000 Blackwell report
Mac mini M4 16 GBToo bigPDF: gpt-oss 20B on Mac mini M4 16 GB report
Mac mini M4 24 GBMXFP4128K34 tok/s12 sPDF: gpt-oss 20B on Mac mini M4 24 GB report
Mac mini M4 32 GBMXFP4128K34 tok/s12 sPDF: gpt-oss 20B on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GBMXFP4128K67 tok/s5.8 sPDF: gpt-oss 20B on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GBMXFP4128K67 tok/s5.8 sPDF: gpt-oss 20B on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GBMXFP4128K67 tok/s5.8 sPDF: gpt-oss 20B on Mac mini M4 Pro 64 GB report
M4 Max 128 GBMXFP4128K104 tok/s2.9 sPDF: gpt-oss 20B on M4 Max 128 GB report
M5 Max 128 GBMXFP4128K112 tok/s0.8 sPDF: gpt-oss 20B on M5 Max 128 GB report
M3 Ultra 96 GBMXFP4128K122 tok/s1.6 sPDF: gpt-oss 20B on M3 Ultra 96 GB report
M3 Ultra 256 GBMXFP4128K122 tok/s1.6 sPDF: gpt-oss 20B on M3 Ultra 256 GB report
M3 Ultra 512 GBMXFP4128K122 tok/s1.6 sPDF: gpt-oss 20B on M3 Ultra 512 GB report
M5 Ultra 512 GBMXFP4128K150 tok/s0.5 sPDF: gpt-oss 20B on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GBMXFP4128K53 tok/s1.7 sPDF: gpt-oss 20B on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GBMXFP4128K53 tok/s1.7 sPDF: gpt-oss 20B on Ryzen AI Max+ 395 128 GB report
RTX SparkMXFP4128K73 tok/s0.6 sPDF: gpt-oss 20B on RTX Spark report
DGX SparkMXFP4128K67 tok/s0.6 sPDF: gpt-oss 20B on DGX Spark report
Dual DGX SparkMXFP4128K67 tok/s0.3 sPDF: gpt-oss 20B on Dual DGX Spark report
DGX Station GB300MXFP4128K511 tok/s0.1 sPDF: gpt-oss 20B on DGX Station GB300 report
2× RTX PRO 6000 BlackwellMXFP4128K287 tok/s0.1 sPDF: gpt-oss 20B on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 BlackwellMXFP4128K287 tok/s0.1 sPDF: gpt-oss 20B on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 BlackwellMXFP4128K287 tok/s0.1 sPDF: gpt-oss 20B on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXMMXFP4128K309 tok/s0.4 sPDF: gpt-oss 20B on A100 80 GB SXM report
H100 SXMMXFP4128K395 tok/s0.1 sPDF: gpt-oss 20B on H100 SXM report
4× H200 NVLMXFP4128K454 tok/s0.1 sPDF: gpt-oss 20B on 4× H200 NVL report
8× H200 NVLMXFP4128K454 tok/s0.1 sPDF: gpt-oss 20B 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:20b
llama.cpp
llama-server -hf unsloth/gpt-oss-20b-GGUF
vLLM
vllm serve openai/gpt-oss-20b
MLX (Mac)
mlx_lm.chat --model mlx-community/gpt-oss-20b-MXFP4-Q4

Questions

How much VRAM does gpt-oss 20B need?

gpt-oss 20B needs 11.8 GB for its weights at 4-bit, 12.9 GB with an 8K context and 13.5 GB with 32K.

What is the minimum VRAM for gpt-oss 20B?

About 10.6 GB at 3-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 RTX 3060 12 GB.

How big is gpt-oss 20B?

20.9B parameters (3.6B active per token): 11.3 GB of weights as released (MXFP4), 11.8 GB at 4-bit.

Can gpt-oss 20B run on an RTX 4090?

Yes. At MXFP4 it holds 128K of context and writes about 197 tokens a second.

More models