models · openai

GPT OSS 120B

OpenAI's 120B open-source model released under Apache-2.0. Strong general reasoning and coding.

What this registry has checked

Quarantine. We hold the publisher’s own published hash for these weights and have recorded it against this entry. We have not downloaded and hashed the weights ourselves, so the hash is the publisher’s claim, not our measurement.

Weight file SHA-256	fee26c4f4e06ca8f9dcb5b135eaeacb732eb10eaae91b5076e2b7fabdf4336e6
Method	hf-published-sha256

Re-check this yourself → Source → Full record →

What is recorded

Publisher	openai
Origin	United States
Licence	Apache-2.0
Parameters	116,829,156,672 (116.8B)
Context window	131,072 tokens
Modality	text->text
Architecture	gpt_oss
Layers	36
Hidden size	2880
Attention	GQA (64 heads, 8 KV)
Native precision	U8
Pulls recorded here	156,019
Downloads reported upstream	5,378,443

Reported scores

MMLU-Pro	80.8
GPQA Diamond	78.2
AIME 2025	93.4
LiveCodeBench	87.8
IFBench	69
Humanity's Last Exam	19.6

Reported by a third party and recorded here, not re-run by us. Source: artificialanalysis.ai/api/v2.

Measured on real hardware

None of these runs are ours. Each figure is the reporter’s, linked to their own evidence, under their own trust label. The number shown is single-stream decode — one user, concurrency 1 — because aggregate throughput across many users is a different quantity and reads as far faster than anyone will see.

Reported to local-ai-registry

Hardware	Engine	Build measured	Decode, 1 stream	Label	Reported by
Intel Arc Pro B70 ×2 devices	llama.cpp	MXFP4	59.4 tok/s @ 2,048	candidate	localmaxxing
NVIDIA DGX Spark GB10	vllm	MXFP4	34.4 tok/s @ 32,768	candidate	localmaxxing
Measured outside 2K–32K context — decode speed depends heavily on context, so these are not comparable with the rows above
RTX PRO 6000 Blackwell	llama.cpp	F16	203 tok/s @ 845	candidate	localmaxxing
Radeon AI PRO R9700 ×3 devices	llama.cpp	F16	72.8 tok/s @ 845	candidate	localmaxxing
Ryzen AI Max+ 395	llama.cpp	F16	38.8 tok/s @ 845	candidate	localmaxxing

From local-ai-registry (MIT), commit 124959d: 6 runs across 5 machines, 0 validated — model revision and runtime pinned, launch accepted — and 6 candidate, their label for useful evidence without a reproducible-launch promise.

Measured by local.ai

Hardware	Build measured	Decode @ 8K	Memory @ 8K	Decode mode
NVIDIA RTX PRO 6000 Blackwell 96GB	—	137 tok/s	90.9 GB	not stated
NVIDIA RTX PRO 6000 Blackwell 96GB	—	128 tok/s	94.1 GB	not stated
NVIDIA RTX PRO 6000 Blackwell 96GB	—	124 tok/s	94.1 GB	not stated
MacBook Pro M5 Max 128GB · 40-core GPU	MXFP4	71.3 tok/s	64.8 GB	not stated
Mac Studio M3 Ultra 96GB · 80-core GPU	MXFP4	67.9 tok/s	64.8 GB	not stated
MacBook Pro M4 Max 128GB · 40-core GPU	MXFP4	62.1 tok/s	64.8 GB	not stated
NVIDIA DGX Spark 128GB	—	33.7 tok/s	102.8 GB	not stated
NVIDIA DGX Spark 128GB	—	33.0 tok/s	88.7 GB	not stated

Measured by local.ai, not by us: 16 runs, 8 more not listed. Each names its engine pinned by image digest and the exact serve command, in the full record.

Elsewhere in the registry

All models → Browse with filters → What will run on my hardware? →

Registry record as of 2026-08-13. Identity, licence and parameter count are properties of a release and do not move; tier can, and the verify link above re-checks it against the live registry rather than this page.