Skip to content

Head-to-head · gpt-oss

gpt-oss 20B vs gpt-oss 120B

gpt-oss 20B needs ~13.2 GB at Q4_K_M; gpt-oss 120B needs ~62.4 GB. That ~49.2 GB gap decides which hardware runs each; they differ on 4 of the 14 devices below.

Spec gpt-oss 20B gpt-oss 120B
Parameters21B MoE117B MoE
Memory at Q4_K_M~13.2 GB~62.4 GB
Context window128k128k
LMArenanot rankednot ranked
LicenseApache-2.0Apache-2.0

Which devices run each

A representative spread across the memory range. Tap a verdict for the full breakdown.

Size at each quantization

Quant gpt-oss 20B gpt-oss 120B
Q2_K 8.8 GB* 49 GB*
Q3_K_M 10.3 GB* 57.2 GB*
Q4_K_M 11.28 GB 59.03 GB
Q5_K_M 15 GB* 83.4 GB*
Q6_K 17.2 GB* 95.9 GB*
Q8_0 0.86 GB 0.79 GB
FP16 42 GB* 234 GB*

* derived from bits-per-weight; unstarred sizes are measured GGUF files.

Bottom line

The larger gpt-oss 120B needs ~62.4 GB; gpt-oss 20B runs on lighter hardware at ~13.2 GB with more headroom and faster responses. Check each against your exact device: what runs gpt-oss 20B · what runs gpt-oss 120B.

FAQ

What is the difference in memory between gpt-oss 20B and gpt-oss 120B?

At Q4_K_M, gpt-oss 20B needs about 13.2 GB and gpt-oss 120B needs about 62.4 GB, a difference of ~49.2 GB.

Should I run gpt-oss 20B or gpt-oss 120B?

Run gpt-oss 120B if your hardware has the ~62.4 GB it needs and you want maximum quality; run gpt-oss 20B (~13.2 GB) for lighter hardware, faster responses, and more memory headroom.

Full breakdowns: gpt-oss 20B · gpt-oss 120B · all models × devices.

Sources

Memory figures are estimates at Q4_K_M, validated 2026-08-03. See methodology.