gpt-oss-20b

21B parametersMoE
reasoningcodingmoe
Parameters
21B
3.6B active
Context Length
32K
Architecture
gpt
Release Date
2025

Available Quantizations

QuantizationFile SizeQuality
Q8_022.4 GBHigh
Q6_K17.3 GBHigh
Q5_K_M15.2 GBGood
Q4_K_M12.8 GBGood
Q3_K_M10.9 GBLower

Can it run? (Q4_K_M quantization)

GeForce RTX 4090
Runs great
~97 tok/s • 24GB VRAM
GeForce RTX 4070
Runs slowly
~10 tok/s • 12GB VRAM
GeForce RTX 3060 12GB
Runs slowly
~7 tok/s • 12GB VRAM
Radeon RX 7900 XTX
Runs great
~92 tok/s • 24GB VRAM
Check with your specific GPU →