GPT-OSS 20B quants compared

28 GGUF builds by real file size, probed from bartowski/openai_gpt-oss-20b-GGUF on Hugging Face (2026-09-02). 21B params, 3.6B active.

Download GPT-OSS 20B Q4_K_M (10.87 GB) — it fits 24 GB of memory with 16k context. With Ollama: ollama run gpt-oss:20b

Every GPT-OSS 20B quant by real file size

QuantWeights+ KV (16k)TotalFits comfortably inQuality
BF1612.85 GB4.0 GB16.9 GB24 GBFull precision (lossless)
MXFP411.28 GB4.0 GB15.3 GB24 GBNative MoE format
Q8_011.28 GB4.0 GB15.3 GB24 GBNear-lossless
Q6_K11.21 GB4.0 GB15.2 GB24 GBExcellent
Q6_K_L11.21 GB4.0 GB15.2 GB24 GBExcellent
Q5_K_L11.09 GB4.0 GB15.1 GB24 GBVery high
Q5_K_M10.92 GB4.0 GB14.9 GB24 GBVery high
Q5_K_S10.92 GB4.0 GB14.9 GB24 GBVery high
Q4_K_M *10.87 GB4.0 GB14.9 GB24 GBHigh — the default pick
Q4_K_L11.07 GB4.0 GB15.1 GB24 GBHigh
Q4_K_S10.87 GB4.0 GB14.9 GB24 GBHigh
Q4_110.8 GB4.0 GB14.8 GB24 GBHigh
IQ4_NL10.77 GB4.0 GB14.8 GB24 GBHigh
IQ4_XS10.77 GB4.0 GB14.8 GB24 GBHigh
Q4_010.73 GB4.0 GB14.7 GB24 GBHigh
Q3_K_XL10.97 GB4.0 GB15.0 GB24 GBAcceptable — visible loss
IQ3_M10.77 GB4.0 GB14.8 GB24 GBAcceptable — visible loss
IQ3_XS10.77 GB4.0 GB14.8 GB24 GBAcceptable — visible loss
IQ3_XXS10.77 GB4.0 GB14.8 GB24 GBAcceptable — visible loss
Q3_K_M10.77 GB4.0 GB14.8 GB24 GBAcceptable — visible loss
Q3_K_S10.76 GB4.0 GB14.8 GB24 GBAcceptable — visible loss
Q3_K_L10.7 GB4.0 GB14.7 GB24 GBAcceptable — visible loss
Q2_K_L11.03 GB4.0 GB15.0 GB24 GBExperimental — not ranked — never recommended
Q2_K10.77 GB4.0 GB14.8 GB24 GBExperimental — not ranked — never recommended
IQ2_M10.75 GB4.0 GB14.8 GB24 GBExperimental — not ranked — never recommended
IQ2_S10.75 GB4.0 GB14.8 GB24 GBExperimental — not ranked — never recommended
IQ2_XS10.72 GB4.0 GB14.7 GB24 GBExperimental — not ranked — never recommended
IQ2_XXS10.72 GB4.0 GB14.7 GB24 GBExperimental — not ranked — never recommended

* default pick. Weights = real GGUF file sizes from bartowski/openai_gpt-oss-20b-GGUF (probed 2026-09-02). KV = fp16 estimate; a q8_0 cache roughly halves it. "Comfortable" = weights + KV within 90% of memory.

Best GPT-OSS 20B quant by memory

MemoryRecommended quantTotal (16k ctx)
24 GBBF1616.9 GB

Why we don't rank GPT-OSS 20B's 2-bit quants

Quants at 2 bits per weight or below (Q2_K, IQ2, IQ1, TQ1) cut file size by roughly half versus Q4, but the quality collapse is steep and non-linear: perplexity spikes, instruction-following degrades, and hallucinations rise. A model that answers faster but wrong is not a smaller model — it is a worse one. ModelFit lists these builds for completeness but never ranks or recommends them.

Run GPT-OSS 20B on your GPU

RTX 3060
12 GB — verdict
RTX 4060
8 GB — verdict
RTX 4060 Ti
16 GB — verdict
RTX 4070
12 GB — verdict
RTX 4070 Ti SUPER
16 GB — verdict
RTX 3090
24 GB — verdict
RTX 4090
24 GB — verdict
RTX 5070 Ti
16 GB — verdict
RTX 5080
16 GB — verdict
RTX 5090
32 GB — verdict

Frequently asked questions

What is the best quantization of GPT-OSS 20B?

Q4_K_M is the default pick: 10.87 GB of weights, high — the default pick quality, fitting comfortably in 24 GB of memory (weights + 16k context KV-cache). Go Q6_K or Q8_0 if you have headroom.

How much memory does GPT-OSS 20B need?

At Q4_K_M, GPT-OSS 20B needs 10.87 GB for the weights plus ~4.0 GB of KV-cache at 16k context — about 14.9 GB total, so a 24 GB card or Mac (90% usable budget) runs it comfortably.

Should I use a Q2_K or IQ2 quant of GPT-OSS 20B?

No. GPT-OSS 20B at 2 bits per weight is a visibly worse model — quality collapse at that bitrate is steep, not gradual. If only a 2-bit build fits your memory, run a smaller model at Q4_K_M instead. ModelFit lists these builds but never recommends them.

Cite this page

ModelFit: GPT-OSS 20B quantization comparison (real GGUF file sizes).
https://modelfit.io/quant-compare/gpt-oss-20b/ (data probed 2026-09-02, CC BY 4.0).