GPT-OSS 120B
Active Parameters
117B
Context Length
128K
Modality
Text
Architecture
Mixture of Experts
KV Cache Rate
72 KB/tok
License
Apache 2.0
Release Date
5 Aug 2025
Knowledge Cutoff
Jun 2024
API Pricing (per 1M)
Input: $0.15 · Output: $0.60
Hardware & System Requirements
Precision
Context
Workload Memory: 230.49 GB
FP16 · 8K context
Base Weights: 227.43 GB
KV Cache: 0.56 GB
Runtime & Graphs: 2.50 GB
Minimum Required Hardware
8x NVIDIA H100 80GB
230.5 / 576.0 GB usable · TP 8
Determine requirements for your specific workload:
Open in VRAM Calculator
Runtime requirements
Checkpoint format
FP16
CUDA
>= 11.8
PyTorch
>= 2.1
Transformers
>= 4.55.0.dev0
| Architecture | Precision | Attention kernel |
|---|---|---|
Blackwell, Hopper | Native FP16 | FlashAttention-3 |
Ada Lovelace, Ampere | Native FP16 | FlashAttention-2 |
Architecture Diagram
Evaluation Benchmarks
Rank
#148
| Benchmark | Score | Rank |
|---|---|---|
Professional Knowledge | high 0.807 | 28 |
Software Engineering | 0.26 | 37 |
Graduate-Level QA | high 0.809 Standard 0.801 | 64 66 |
Agentic Index | high 0.04 | 103 |
Coding Index | high 0.30 low 0.21 | 119 131 |
General Text | 1352 | 134 |
Intelligence Index | high 0.04 | 333 |
Summarization Archived | 0.976 | 🥈 2 |
General Knowledge Reference | 0.9 | 🥈 2 |
Coding Archived | high 0.42 | 17 |
Rankings
Overall Rank
#148
Coding Rank
#126
About GPT-OSS 120B
GPT-OSS 120B is a large open-weight model from OpenAI, designed to operate in data centers and on high-end desktops and laptops. It is developed to support advanced reasoning, agentic tasks, and diverse developer use cases, functioning as a text-only model for both input and output modalities.
Specifications
Attention
Attention Structure
Multi-Head Attention
Attention Heads
64
Key-Value Heads
8
Attention Head Dimension
64
Position Embedding
Absolute Position Embedding
RoPE Theta
150,000
Sliding Window Attention
Yes
Sliding Window Size
128
Normalization
RMS Normalization
Activation Function
SwigLU
Dimensions
Base Precision
FP16
Hidden Dimension Size
2,880
Number of Layers
36
FFN Intermediate Size (Dense)
2,880
Tokenizer
Vocabulary Size
201,088
Mixture of Experts
Total Expert Parameters
5.1B
Number of Experts
128
Active Experts
4