How much VRAM does gpt-oss 20B (MoE) need?

gpt-oss 20B (MoE) requires approximately 0.0 GB VRAM at Q4_K_M quantization with 1K context and 10% overhead. At Q8_0 it needs ~0.0 GB, and at FP16 ~0.0 GB. Use the PCPARTGUIDE VRAM Calculator for exact requirements at any context length and overhead setting.

Can I run gpt-oss 20B (MoE) on a consumer GPU?

Yes, gpt-oss 20B (MoE) can run on consumer GPUs. At Q4_K_M quantization with moderate context length, it fits on a 24 GB GPU. For best performance, use a 24 GB or larger GPU with at least 8 GB VRAM for reasonable context lengths.

What is gpt-oss 20B (MoE)'s context window?

gpt-oss 20B (MoE) supports a maximum context window of 131,072 tokens. The default context is 16,384 tokens. Larger contexts require proportionally more VRAM for the KV cache — use the VRAM Calculator to estimate exact requirements at your target context length.

What quantization level should I use for gpt-oss 20B (MoE)?

For gpt-oss 20B (MoE), start with Q4_K_M (~0.0 GB) if VRAM-limited. Q8_0 offers better quality at ~2× the VRAM. If you have enough VRAM, Q6_K (~1.5× Q4 VRAM) provides a good balance. For models this large, Q3_K_M may be necessary to fit on consumer hardware.

How does gpt-oss 20B (MoE) compare to similar models?

gpt-oss 20B (MoE) is a 21.0B MoE model from the OpenAI family. Browse other OpenAI models on the PCPARTGUIDE directory for comparable alternatives with different parameter counts and VRAM requirements. Use the Model Transparency scores to assess documentation quality and reproducibility.

OpenAIMoEApache 2.0

gpt-oss 20B (MoE)

21.0B

Parameters

3.6B

Active

128K

Max Context

MoE

Architecture

—

Released

Text

Modality

About gpt-oss 20B (MoE)

gpt-oss 20B (MoE) is a mixture-of-experts (MoE) transformer language model from the OpenAI family, containing 21B parameters across 24 layers. It has 21B total parameters loaded into VRAM with 3.6B active per token. It supports up to 131K tokens of context with a hidden dimension of 2880 and 8 KV heads for efficient grouped-query attention (GQA). Apache 2.0. MoE: 32 experts, top-4 routing. Fits 16GB at MXFP4. Strong local reasoning.

Reasoning

Technical Specifications

Total Parameters21.0B

Active Parameters3.6B per token

ArchitectureMixture of Experts

Total Experts3.6

Attention TypeGQA (MoE)

Hidden Dimensiond = 2,880

Transformer Layers24

Attention Heads64

KV Headsn_kv = 8

Head Dimensiond_head = 64

Activation FunctionSwiGLU

NormalizationRMSNorm

Position EmbeddingRoPE

System Requirements

Estimated VRAM at 10% overhead for different quantization methods and context sizes.

Quantization	1K ctx	128K ctx
Q4_K_M0.50 B/W ~97% of FP16	10.90Consumer GPU	16.85Consumer GPU
Q8_01.00 B/W ~100% of FP16	21.76Consumer GPU	27.71Datacenter GPU
F162.00 B/W Reference	43.47Datacenter GPU	49.42Datacenter GPU

Fits 24 GB consumer GPU

Fits 80 GB datacenter GPU

Requires cluster / multi-GPU

Other OpenAI Models

View All

gpt-oss 120B (MoE)

Params

117.0B

Layers

Context

128K

5.1B active

Find the right GPU for gpt-oss 20B (MoE)

Use the interactive VRAM Calculator to see exactly how much memory you need at any quantization level, context length, and overhead setting.

Open VRAM Calculator Best GPU Guide VRAM Guide