gpt-oss-120b & gpt-oss-20b Model Card
by OpenAI
OpenAI's official model card for its first open-weight (Apache 2.0) reasoning models — architecture, training, and agentic capabilities.
Overview
Published by OpenAI on arXiv (August 2025), the gpt-oss-120b & gpt-oss-20b Model Card documents OpenAI's first open-weight language models since GPT-2. Both are efficient mixture-of-experts (MoE) transformers trained with large-scale distillation and reinforcement learning, and released — weights, inference implementations, tool environments, and tokenizers — under a permissive Apache 2.0 license. The card details their architecture, a rendered chat format for clear instruction following and role delineation, and native agentic capabilities including web browsing, Python tool execution, and developer-provided function calling. gpt-oss-120b reaches near-parity with o4-mini on core reasoning benchmarks while running on a single 80 GB GPU, and gpt-oss-20b runs on edge devices with 16 GB of memory. It also reports mathematics, coding, and safety benchmark results, making it a primary source for engineers choosing an open-weight reasoning model.
At a Glance
- Topic
- Models
- Level
- Advanced
- Format
- Paper
- Cost
- Free
- Duration
- ~30-page technical report
- Provider
- OpenAI
- Hands-on
- No
- Certificate
- None
What You’ll Learn
- ✓How OpenAI's open-weight gpt-oss models are architected (mixture-of-experts transformers)
- ✓The training recipe: large-scale distillation plus reinforcement learning
- ✓Native agentic features: browsing, Python execution, and function/tool calling
- ✓Reasoning, math, coding, and safety benchmark results for both model sizes
- ✓Hardware footprint and trade-offs (single 80 GB GPU vs. 16 GB edge device)
- ✓What the Apache 2.0 release enables for self-hosted and fine-tuned deployments
Highlights
- •Primary source: OpenAI's own model card, not a third-party summary
- •Covers OpenAI's first open-weight models since GPT-2, under Apache 2.0
- •Concrete benchmark and hardware numbers for real deployment decisions
- •Details agentic tool-use behavior relevant to agent builders
Who It’s For
Best For
- ✓AI engineers evaluating open-weight reasoning models for production
- ✓Teams planning self-hosted or on-device LLM deployments
- ✓Practitioners fine-tuning or serving open models who need architecture details
Prerequisites
- •Familiarity with transformer / LLM architecture concepts
- •Understanding of model benchmarks and inference/serving basics
FAQ
What is gpt-oss-120b & gpt-oss-20b Model Card?
The official OpenAI model card for gpt-oss-120b and gpt-oss-20b, two open-weight mixture-of-experts reasoning models released under Apache 2.0 in August 2025. Essential reading for AI engineers evaluating open-weight models for agentic, tool-using, and locally-served workloads.
Is gpt-oss-120b & gpt-oss-20b Model Card free?
gpt-oss-120b & gpt-oss-20b Model Card is free to access.
What level is gpt-oss-120b & gpt-oss-20b Model Card for?
gpt-oss-120b & gpt-oss-20b Model Card is aimed at a advanced audience. Recommended background: Familiarity with transformer / LLM architecture concepts, Understanding of model benchmarks and inference/serving basics.
How long does gpt-oss-120b & gpt-oss-20b Model Card take?
Expect roughly ~30-page technical report. Most learners work through it at their own pace.
What will I learn from gpt-oss-120b & gpt-oss-20b Model Card?
You'll learn: How OpenAI's open-weight gpt-oss models are architected (mixture-of-experts transformers); The training recipe: large-scale distillation plus reinforcement learning; Native agentic features: browsing, Python execution, and function/tool calling; Reasoning, math, coding, and safety benchmark results for both model sizes; Hardware footprint and trade-offs (single 80 GB GPU vs. 16 GB edge device); What the Apache 2.0 release enables for self-hosted and fine-tuned deployments.