AI Index

gpt-oss-120b

Language OpenAI Verified Released Aug 5, 2025
  • In Text Out Text
  • Reasoning· low, medium, high
  • Tool use
  • Structured output
  • Open weights

Prices

1 offering · prices as listed
Cloudflare Workers AI Global · USD @cf/openai/gpt-oss-120b
Conditions/ 1M tokensInputOutput
Standard $0.35$0.75

Prices from Cloudflare Workers AI pricing (checked Oct 31, 2026)

Standard: the price a request with no conditions pays (the headline price elsewhere on the site).

Details

Knowledge cutoff
Jun 2024
Licence
Apache 2.0
Family
gpt-oss
Model ID

Not stated on OpenAI’s pages: max input, reasoning can be off, max thinking budget, aliases.

Run locally

Weights: openai/gpt-oss-120b
  • Parameters 116.8B
  • Active 5.1B · 128 experts, 4 per token
  • Native precision MXFP4
  • Layers 36
  • KV cache per token 72 KB
  • Config positions 128K (config.json max_position_embeddings)
  • API context 128K (maker’s API)
My hardware GeForce RTX 4090 24 GB · 1008 GB/s · 64 GB DDR5 dual channel · 89.6 GB/s
On your hardware · GeForce RTX 4090
Needs system memory MXFP4 (native) (automatic) 8K context 21–32 tok/s26/36 layers in system memory
WeightsKV cacheOverheadIn system memoryUsable
Needs
≈ 64.1 GB (weights 60.8 (maker’s file) + KV cache 0.3 + overhead 3.0)
Usable
21.6 GB (90% of its memory)
In system memory
43.9 GB · 10/36 layers on the GPU (--n-gpu-layers 10)
Generation speed
21–32 tok/s (reads the active 5.1B per token)
First token
about 0.16–0.25 s for a 2K prompt
PrecisionWeightsNeeds at 8KOn your hardwareLongest context
MXFP4 (native)maker’s file60.8 GB≈ 64.1Needs system memory 21–32 tok/s—
Q3_K_Mquality drops53.2 GB≈ 56.1Needs system memory 26–39 tok/s—
IQ3_XXSquality drops41.6 GB≈ 44.0Needs system memory 39–58 tok/s—
Q2_Kquality drops40.8 GB≈ 43.1Needs system memory 41–62 tok/s—
IQ2_XXSquality drops28.0 GB≈ 29.7Needs system memory 93–139 tok/s—

Click a row to choose a precision. “Maker’s file” sizes are the maker’s own files; others are parameters × bits per weight.

Compared with other hardware

Each at the highest precision it runs. Click one for details.

Smallest that runs it (MXFP4 (native), 8K): GPU NVIDIA H100 SXM; unified memory Apple M5 Ultra (64-core GPU) · 96 GB.

Run commands

vLLM
vllm serve openai/gpt-oss-120b --max-model-len 8192
Memory to fine-tune
Full≈ 1743.5 GBDoesn’t fit your hardware
LoRA≈ 220.4 GBDoesn’t fit your hardware
QLoRA≈ 64.0 GBDoesn’t fit your hardware

Batch 1, 2,048-token sequences, gradient checkpointing; LoRA and QLoRA train rank-16 adapters on the attention projections.

Estimates; facts from the maker’s Hugging Face repositories, as of 2026-11-06. How it’s estimated

Cited scores

From third-party leaderboards, not measured by us
  • Epoch Capabilities Index 140.2

    #97 of 118 listed models · CI 138.4–142

  • LMArena Text 1,352

    #125 of 163 listed models · CI 1,347–1,356 · 30,200 votes

  • Epoch AI · Mystery Game Puzzles 2%

    #77 of 77 listed models · medium effort · ±1.41

    Also listed: high effort 0%

  • EQ-Bench Creative Writing v3 956

    #82 of 93 listed models

  • Epoch AI, 'Capabilities & benchmarking'. Published online at epoch.ai. Retrieved from 'https://epoch.ai/benchmarks'. Licensed CC BY 4.0. · License · as of Oct 28 – Nov 5, 2026
  • LMArena (Arena), leaderboard-dataset on Hugging Face, licensed CC BY 4.0 · License · as of Oct 28, 2026
  • EQ-Bench by Samuel J. Paech (eqbench.com; EQ-bench/EQ-bench-site on GitHub), MIT License. · License · as of Oct 28, 2026
  • changes we made

Changes

All changes →
  1. Now priced on Cloudflare Workers AI Input $0.35 · Output $0.75

Sources

OpenAI’s official policies: data use, privacy, terms →

OpenAI · United States · openai.com

Snapshots, the same family and other models from this provider