Introducing gpt-oss-20bOpenAIReleased September 4, 2025

gpt-oss-20b

Medium-sized Apache 2.0 open-weight model with 21B parameters (3.6B active), designed for local execution on consumer GPUs.

reasoning-llmOpen WeightsFree Open Weights (Consumer GPU)Context: 131.072K (131,072 tokens)Arena ELO: 1310

Technical Specifications

Architecture Type
Lightweight Sparse MoE Reasoning Model
Total Parameters
21B
Active Parameters (MoE)
3.6B
Context Window
131.072K (131,072 tokens)
Max Output Tokens
32.768K (32,768 tokens)
Knowledge Cutoff
October 2024
Supported Modalities
text, code, reasoning
License & Access
Apache 2.0
Weights Formats
safetensors, gguf, fp8, awq

Benchmark Evaluations

Math500
81.2
Mmlu Pro
68.9
Humaneval
86.4
Chatbot Arena ELO
1310

Deep Architectural Overview

gpt-oss-20b delivers native reasoning, agentic tool use, and fast local execution on consumer hardware (16GB VRAM). Perfect for private edge deployment, offline analysis, and fine-tuning.

Strengths & Considerations

Core Strengths
  • Runs on consumer GPUs (RTX 4080 / Apple Silicon)
  • Apache 2.0 license with no copyleft restrictions
  • Native reasoning and tool calling support
Known Limitations
  • Smaller knowledge capacity than 120B model

Token & API Pricing

Input Tokens (1M)Free (Open Weights)
Output Tokens (1M)Free (Open Weights)
Self-Hosted Min VRAM16 GB
Pricing is verified directly against OpenAI's developer documentation and API rate sheets.

Similar & Alternative Models

Explore other frontier models from OpenAI and comparable reasoning engines.

Browse all models
OpenAI

OpenAI GPT-6 model for focused, high-volume workloads where cost efficiency is the priority.

1.05M ctxUSD 0.10 / 1M input tok; USD 0.50 / 1M output tok
OpenAI

OpenAI GPT-6 model built for complex coding and agentic workflows, balancing intelligence and cost.

1.05M ctxUSD 2 / 1M input tok; USD 10 / 1M output tok
OpenAI

OpenAI’s premier sixth-generation frontier model, built for the most demanding end-to-end software engineering and scientific research tasks.

1.05M ctx$10.00 / 1M tok (GPT-6 Frontier)