onrup

gpt-oss · OpenAI

Fine-tuning gpt-oss 20B

What does it take to fine-tune gpt-oss 20B?

gpt-oss 20B needs 40 GB for half-precision LoRA training and 24 GB in four-bit, and 24 GB to serve. It supports supervised fine-tuning, under the Apache 2.0 licence. The cheapest qualifying class is L4 at $0.17 per GPU-hour.

OpenAI's smaller open-weight release: twenty-one billion total parameters, 3.6 billion active, Apache-2.0. Twenty-four gigabytes to serve means a mainstream class can host it, which is a better ratio than any dense model of comparable capability. Experimental in this catalogue.

What to know before choosing it

Twenty-four gigabytes to serve a model with twenty-one billion total parameters is the headline: a mainstream class can host it, which is a better ratio than any dense model of comparable capability manages.

Training is where the sparsity stops helping. Forty gigabytes for half precision, and the usual mixture-of-experts difficulties apply. It is marked experimental in this catalogue for that reason rather than because of the model's quality.

Specification

gpt-oss 20B specification
Parameters20.9B3.6B active per token
ArchitectureMixture of experts
Size tierLarge
LoRA training memory40 GBHalf precision, frozen base
QLoRA training memory24 GBFour-bit base, higher-precision adapter
Serving memory24 GBHalf precision, before attention cache
ObjectivesSFT
LicenceApache 2.0
Repositoryopenai/gpt-oss-20b

What it costs to train

Every class with enough memory for four-bit training, cheapest first. Total cost is the rate multiplied by wall time, so the cheapest rate is not always the cheapest run — a faster class that finishes sooner frequently wins.

GPU classVRAMTraining / hrServing / hrFits
L424 GB$0.17$0.20QLoRA only
RTX 309024 GB$0.22$0.26QLoRA only
RTX 409024 GB$0.38$0.43QLoRA only
A4048 GB$0.42$0.48LoRA and QLoRA
RTX 6000 Ada48 GB$0.61$0.71LoRA and QLoRA
A600048 GB$0.65$0.75LoRA and QLoRA
L40S48 GB$0.78$0.90LoRA and QLoRA
A100 40 GB40 GB$1.17$1.35LoRA and QLoRA
A100 80 GB80 GB$1.68$1.94LoRA and QLoRA
H100 80 GB80 GB$2.59$2.99LoRA and QLoRA
H200141 GB$4.55$5.25LoRA and QLoRA

Half-precision LoRA needs 40 GB, so the cheapest class for it is A40 at $0.42 per hour. Below that, training has to be quantised.

Serving fits on L4 at $0.20 per hour — before the attention cache, which grows with context length and concurrency.

Good starting point for

This is a mixture-of-experts model

20.9B total parameters with 3.6B active per token. Memory scales with the total; inference compute scales with the active count. That makes it attractive when memory is cheaper than compute for your workload, and unattractive when the reverse holds. Quoting only one of the two numbers is how these models get misrepresented in both directions.

What mixture of experts means →

Other sizes in this family

ModelParamsQLoRAObjectives
gpt-oss 120B116.8B80 GBSFT

Comparable sizes elsewhere

Frequently asked questions

How much VRAM does it take to fine-tune gpt-oss 20B?

40 GB for half-precision LoRA and 24 GB for four-bit QLoRA. Serving needs 24 GB. Preference tuning roughly doubles the training figure, because a frozen reference model is held alongside the one being trained.

What is the cheapest way to fine-tune gpt-oss 20B?

Four-bit QLoRA on L4 at $0.17 per GPU-hour is the cheapest class that meets the 24 GB threshold. Note that quantised training is slower per step, so a faster class sometimes costs less over the whole run.

Can I download the weights after fine-tuning gpt-oss 20B?

Yes. Every finished run exposes its trained weights for download, and publishing to a model hub is a single call with a generated model card recording the base model and version the adapter applies to.

What licence does gpt-oss 20B carry?

Apache 2.0. The licence follows the fine-tune — a derivative inherits the base model’s terms, and those terms pass to anyone you give the model to.

Last verified 6 August 2026. Memory thresholds are the platform's own admission limits.

Fine-tune gpt-oss 20B

Upload a dataset, forecast the run, and see the cost before any compute is leased.