FitMyLLM
OpenAI/Mixture of Experts

OpenAIGPT-OSS 20B

OpenAI's first open-weight model. MoE with 3.6B active params — runs on 12GB RAM. Matches o3-mini on reasoning benchmarks.

chatcodingreasoningtool_use
21B
Parameters (3.6B active)
128K
Context length
18
Benchmarks
10
Quantizations
Architecture
MoE
Released
2026-02-14
Layers
24
KV Heads
8
Head Dim
64
Family
gpt-oss

Quantization Options

Context length:
QuantBitsVRAM @ 16KQuality
Q3_K_M4
11.3 GB
11.0 + 0.3 KV
low
Q3_K_L4.3
12.1 GB
11.8 + 0.3 KV
moderate
IQ4_XS4.46
12.5 GB
12.2 + 0.3 KV
moderate
Q4_K_S4.67
13.0 GB
12.7 + 0.3 KV
moderate
Q4_K_M4.89
13.6 GB
13.3 + 0.3 KV
good
Q5_K_S5.57
15.4 GB
15.1 + 0.3 KV
good
Q5_K_M5.7
15.7 GB
15.5 + 0.3 KV
good
Q6_K6.56
18.0 GB
17.7 + 0.3 KV
excellent
Q8_08.5
23.1 GB
22.8 + 0.3 KV
lossless
FP1616
42.8 GB
42.5 + 0.3 KV
lossless

Select your GPU above to see speed estimates and compatibility for each quantization.

Deploying for a team or in production? Size GPUs, cost & scaling in Enterprise →
Ready to run this?Rent by the hour

Rent a GPU and run GPT-OSS 20B now

Spin up an A100 / H100 / 4090 in ~60s. Pay by the second. Cancel anytime.

Community Ratings

Loading ratings...

Benchmarks (18)

Run this model

Easiest way to get started·Beginners
Docs ↗
curl -fsSL https://ollama.com/install.sh | sh
$ollama run gpt-oss:20b

Ollama's default build of this model. For Q4_K_M specifically, pick a tag from ollama.com/library/gpt-oss/tags.

Setup guide
>_

Auto-setup with fitmyllm CLI

Detects your GPU, recommends the best model, downloads it, and starts chatting — zero config. Benchmarks your speed and contributes anonymous data to improve predictions.

pip install fitmyllmthen run fitmyllmLearn more
Auto-detect GPULive tok/s in chatSpeed benchmarks9 inference engines

GPUs that can run this model

At Q4_K_M quantization. Sorted by minimum VRAM.

Apple M1 Pro (16GB)
16 GB VRAM • 200 GB/s
APPLE
$999
Apple M2 Pro (16GB)
16 GB VRAM • 200 GB/s
APPLE
$1299
Apple M4 (16GB)
16 GB VRAM • 120 GB/s
APPLE
$499
NVIDIA Tesla T4 16GB
16 GB VRAM • 320 GB/s
NVIDIA
$800
NVIDIA V100 PCIe 16GB
16 GB VRAM • 900 GB/s
NVIDIA
$2000
Apple M1 (16GB)
16 GB VRAM • 68.25 GB/s
APPLE
$699
Apple M2 (16GB)
16 GB VRAM • 100 GB/s
APPLE
$799
Apple M3 (16GB)
16 GB VRAM • 100 GB/s
APPLE
$799
NVIDIA Tesla P100 DGXS
16 GB VRAM • 732 GB/s
NVIDIA
NVIDIA Tesla P100 PCIe 16 GB
16 GB VRAM • 732 GB/s
NVIDIA
NVIDIA Tesla P100 SXM2
16 GB VRAM • 732 GB/s
NVIDIA
NVIDIA Tesla V100 PCIe 16 GB
16 GB VRAM • 897 GB/s
NVIDIA
NVIDIA Tesla V100 SXM2 16 GB
16 GB VRAM • 1130 GB/s
NVIDIA

Find the best GPU for GPT-OSS 20B

Build Hardware for GPT-OSS 20B
Spec sheet

GPT-OSS 20B — 21B MoE.

Specifications
Parameters
21B (3.6B active)
Architecture
Mixture of Experts
Context Length
128K tokens
Capabilities
chat, coding, reasoning, tool_use
Release Date
2026-02-14
Provider
OpenAI
Family
gpt-oss
VRAM requirements
QuantBPWVRAMQuality
Q3_K_M411.0 GB88%
Q3_K_L4.311.8 GB90%
IQ4_XS4.4612.2 GB92%
Q4_K_S4.6712.7 GB93%
Q4_K_M4.8913.3 GB94%
Q5_K_S5.5715.1 GB96%
Q5_K_M5.715.5 GB96%
Q6_K6.5617.7 GB97%
Q8_08.522.8 GB100%
FP161642.5 GB100%
Benchmark scores
HumanEval81.7
MMLU-PRO85.3
IFEval69.5
LiveCodeBench65.2
AIME98.7
MATH-50089.3
GPQA Diamond71.5
HLE10.9
AA Intelligence20.8
AA Coding14.4
AA Math62.3
IFBench57.8
Terminal-Bench Hard4.5
τ²-Bench50.3
SciCode34.0
AA Long Context31.0
EQ-Bench Creative Writing v3665
EQ-Bench 3985
Run command

Run GPT-OSS 20B locally with Ollama — needs 13.3 GB VRAM at Q4_K_M:

$ollama run gpt-oss:20b
Compatible GPUs
30 @ Q4_K_M