Skip to content
local-ai

gpt-oss-120b

OpenAI · Text generation · MoE 117B-A5B · 131k context · Released 5 August 2025

Commercial use permitted Open weights Apple Silicon

gpt-oss-120b was OpenAI's return to open weights, alongside the smaller gpt-oss-20b. Its mixture-of-experts design keeps active parameters low, and it ships natively in MXFP4, which is why a model over 100 billion parameters fits on one 80GB card. It is the accessible end of server-class: bigger than anything a consumer GPU runs, but a single high-end accelerator or a high-memory Mac, rather than a cluster.

Strengths

  • Strong reasoning for an open model, which OpenAI reports as near its o4-mini
  • Runs on a single 80GB GPU, unusual for a model of this size, thanks to native MXFP4
  • Mixture-of-experts keeps generation fast, with only 5.1B parameters active per token
  • Apache 2.0, so no commercial-use conditions

Weaknesses

  • Needs an 80GB GPU or a high-memory Mac, so it is not a consumer-card model
  • Text only, with no image or audio support
  • A reasoning model, so it is verbose and slower than a plain model
  • The reasoning comparison is OpenAI's own figure, so confirm it on your own tasks

Hardware requirements

QuantisationApprox. VRAMNotes
MXFP4 (native)~61GBThe native format, about 61GB, which runs on a single 80GB GPU
BF16~234GBFull precision, multi-GPU territory

Optimised builds available for Apple Silicon.

What you'd need to run this

Roughly what a machine to run this would need, at up to three levels of quality. Memory is the deciding factor.

Minimum to run it

MXFP4 (native) · ~61GB needed

96GB of unified memory

AMD Ryzen AI Max+ 395 (Strix Halo)

or an 80GB-class data-centre card, which is usually rented by the hour, NVIDIA A100 80GB .

Unified memory is shared with the model, so it is already counted above.

around £1,100–£2,000

See the Quiet large-model box build What else 96GB runs →

Best quality

BF16 · ~234GB needed

A multi-GPU server, roughly 4× 80GB-class GPUs

Server memory in the hundreds of gigabytes or more, plus fast interconnect.

tens of thousands of pounds, or rented by the hour

Usually rented — see the cost calculator

Licence

Apache 2.0 — read the licence

Benchmarks

BenchmarkScoreSourceAs of
GPQA Diamond80.1 OpenAI gpt-oss model card (high reasoning, no tools) August 2025
AIME 202495.8 OpenAI gpt-oss model card (high reasoning, no tools) August 2025
AIME 202592.5 OpenAI gpt-oss model card (high reasoning, no tools) August 2025
SWE-bench Verified62.4% OpenAI gpt-oss model card (high reasoning) August 2025
Aider Polyglot41.8% Aider polyglot leaderboard (independent) August 2026
Artificial Analysis Intelligence Index v4.1.124 Artificial Analysis (independent) August 2026

How it compares

How this model’s reported scores sit against other models we cover, on the same benchmarks. This model is highlighted.

SWE-bench Verified

higher is better

Aider Polyglot

higher is better

AIME 2025

higher is better

Artificial Analysis Intelligence Index v4.1.1

higher is better

AIME 2024

higher is better

GPQA Diamond

higher is better

gpt-oss-120b: common questions

What hardware do I need to run gpt-oss-120b?
At its most compressed (MXFP4 (native)) it needs roughly 61GB of VRAM, and about 80GB for good quality. VRAM figures are approximate and depend on context length and settings.
Is gpt-oss-120b free for commercial use?
Yes. gpt-oss-120b is licensed under Apache 2.0, which permits commercial use with no meaningful conditions.
Can I run gpt-oss-120b on Apple Silicon?
Yes. gpt-oss-120b has builds optimised for Apple Silicon, through MLX or GGUF on a Mac.
What is gpt-oss-120b's context window?
gpt-oss-120b has a context window of 131,072 tokens, about 131k.

Availability

Where to get quantised weights

Some of the best quantised weights are made by the community, not the model’s authors. Look this model up on these providers:

  • Unsloth GGUF

    Ships natively in 4-bit (MXFP4), so community builds mostly repackage it for runtimes rather than compress it further.

  • Bartowski GGUF

    A wide, reliable range of imatrix GGUF quants, typically Q2 through Q8.

Recommended for

  • A strong open reasoning model on a single 80GB GPU
  • High-memory Apple Silicon machines with 96GB or more
  • Commercial reasoning work needing a permissive licence

Related models

Run it with

Related guides

Glossary

Our coverage

Catalogue entry last verified 13 August 2026. Specifications change; verify anything you are about to spend money on.