Model

gpt-oss-120b (High)

OPEN

OpenAI · gpt-oss family · released Aug 5, 2025

OpenAI's first open-weight model since GPT-2; 117B MoE (5.1B active), Apache 2.0.

ReasoningCodingVisionFunction callingTool useAgentic

1839.3

Elo · rank #93

Parameters

117B

Active params

5.1B (MoE)

Context

128K tokens

Architecture

Mixture-of-experts transformer, 128 experts/layer, top-4 active, 36 layers

License

Apache 2.0

Languages

—

API price (in/out)

No hosted API

Modalities

text

Benchmark results

Bar shows position within the tracked field; marker = field best

Aider PolyglotCoding44.4%#29

best: Claude Opus 4.5 · 89.4%

AIMEMath95.8%#11

best: GPT-5.2 · 100.0%

CodeforcesCoding2463#8

best: DeepSeek-V4-Pro (Think Max) · 3206

GPQA DiamondReasoning80.1%#86

best: GPT-6 Astra · 96.0%

Humanity's Last ExamReasoning14.9%#81

best: Claude Opus 5 · 64.7%

MMLUKnowledge90.0%#11

best: OpenAI o3 · 92.9%

MMMLUKnowledge81.3%#23

best: Gemini 3.1 Pro · 92.6%

SWE-bench VerifiedCoding62.4%#79

best: Claude Opus 5 · 96.0%

tau-benchAgents67.8%#10

best: Claude Opus 4.1 · 82.4%

Run it locally

VRAM @ Q4

73 GB

VRAM @ FP16

234 GB

Fits on (Q4)

M3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB

Does not fit on a single RTX 4090 (24GB); OpenAI specifies a single 80GB GPU (H100/MI300X) for native MXFP4 inference.

Quantizations

MXFP4 (native) · GGUF Q4_K_M · GGUF Q8_0

Fine-tune it

Permissive

QLoRA79.6 GB1× A100 80GB

LoRA249.2 GB2× H200 141GB

Full fine-tune1877.8 GBbeyond 8× B200

QLoRA SFT on ~10k samples ≈ $3.59 (1× A100 80GB)

Plan a fine-tune →

gpt-oss family

Elo progression across releases

gpt-oss-120b (High)Aug 20251839.3 gpt-oss-120b (Low)Aug 20251291.7 gpt-oss-120b (Medium)Aug 20251533.5 gpt-oss-20b (High)Aug 20251527.6 gpt-oss-20b (Low)Aug 20251040.0 gpt-oss-20b (Medium)Aug 20251340.3

Resources

HFopenai/gpt-oss-120b GHopenai/gpt-oss DOCScookbook.openai.com APIAvailable via OpenAI API

API price weights · each benchmark row carries its own source badge (see methodology)