GPT-OSS Models
Open-weight reasoning models with Apache 2.0 license
Explore the technical specifications, capabilities, and deployment options for GPT-OSS 120B and 20B models.
Model Overview
GPT-OSS 120B
Flagship
Our flagship model with 120 billion parameters, delivering state-of-the-art performance on reasoning, coding, and mathematics.
120B Parameters
Parameters
131k Context Window
Context
Apache 2.0
License
GPT-OSS 20B
Edge Optimized
Optimized for edge deployment with 20 billion parameters, maintaining excellent performance with lower resource requirements.
20B Parameters
Parameters
131k Context Window
Context
Apache 2.0
License
Model Architecture
Both models feature advanced Mixture of Experts (MoE) architecture with cutting-edge capabilities.
Mixture of Experts
128 experts with Top-4 routing for efficient computation
Extended Context
131k token context window using RoPE technology
Chain-of-Thought
Advanced reasoning capabilities with step-by-step problem solving
Tool Integration
Native support for function calling and tool use
Model Capabilities
Comprehensive evaluation across multiple domains and benchmarks.
| Benchmark | Description | GPT-OSS 120B | GPT-OSS 20B |
|---|---|---|---|
| MMLU | Massive Multitask Language Understanding | 85% | 78% |
| Codeforces | Competitive programming challenges | 2622 | 2516 |
| AIME 2024 | American Invitational Mathematics Examination | Excellent | Strong |
| HealthBench | Medical and healthcare evaluations | Superior | Excellent |
Deployment Options
Multiple deployment options to suit your infrastructure needs.
Ollama
Simple local deployment with one-command setup
ollama run gpt-oss:120bVLLM
High-performance inference for production workloads
vllm serve openai/gpt-oss-120bHugging Face
Direct integration with transformers library
from transformers import AutoModelForCausalLMDocker
Containerized deployment for cloud environments
docker run -p 8000:8000 gpt-oss:120bStart Using GPT-OSS Models
Ready to deploy? Explore our projects showcase to see real-world implementations.