onrup

gpt-oss · OpenAI

Fine-tuning gpt-oss 120B

What does it take to fine-tune gpt-oss 120B?

gpt-oss 120B needs 220 GB for half-precision LoRA training and 80 GB in four-bit, and 80 GB to serve. It supports supervised fine-tuning, under the Apache 2.0 licence. The cheapest qualifying class is A100 80 GB at $1.68 per GPU-hour.

A hundred and seventeen billion parameters with 5.1 billion active. A single 80GB GPU will serve it; full half-precision fine-tuning needs far more, so the practical route is a four-bit run at 80GB. The largest Apache-2.0 model in the catalogue.

What to know before choosing it

A hundred and seventeen billion parameters that serve on a single 80GB class is the most striking number in the catalogue, and it is entirely a consequence of only 5.1 billion being active per token.

Full half-precision fine-tuning needs far more than any single class provides, so the practical route is a four-bit run at 80GB. It is also the largest Apache-2.0 model here, which for some organisations is the deciding factor on its own.

Specification

gpt-oss 120B specification
Parameters116.8B5.1B active per token
ArchitectureMixture of experts
Size tierExtra large
LoRA training memory220 GBHalf precision, frozen base
QLoRA training memory80 GBFour-bit base, higher-precision adapter
Serving memory80 GBHalf precision, before attention cache
ObjectivesSFT
LicenceApache 2.0
Repositoryopenai/gpt-oss-120b

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
A100 80 GB80 GB$1.68$1.94QLoRA only
H100 80 GB80 GB$2.59$2.99QLoRA only
H200141 GB$4.55$5.25QLoRA only

Serving fits on A100 80 GB at $1.94 per hour — before the attention cache, which grows with context length and concurrency.

Good starting point for

This is a mixture-of-experts model

116.8B total parameters with 5.1B 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 20B20.9B24 GBSFT

Comparable sizes elsewhere

Frequently asked questions

How much VRAM does it take to fine-tune gpt-oss 120B?

220 GB for half-precision LoRA and 80 GB for four-bit QLoRA. Serving needs 80 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 120B?

Four-bit QLoRA on A100 80 GB at $1.68 per GPU-hour is the cheapest class that meets the 80 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 120B?

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 120B 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 120B

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