gpt-oss (OpenAI): an Apache-licensed reasoning model for one GPU

Contents

1. What it does for a company
2. Strengths and limits
3. Pricing
4. Questions to ask before choosing gpt-oss
5. Sources

Type	Open-weight model family
Pricing	Published
Website	huggingface.co

gpt-oss is OpenAI’s open-weight family: two text models published on Hugging Face in August 2025 under Apache 2.0, with no newer version since⁠1openai/gpt-oss-120bOpenAI (Hugging Face).

Best for: A capable reasoning model that fits on one 80 GB GPU, under Apache 2.0.

Not the right fit for: Teams that want a current-generation model: gpt-oss has not been updated since August 2025.

What it does for a company

gpt-oss-120b has 117 billion parameters, 5.1 billion active per token, and fits on a single 80 GB GPU. gpt-oss-20b has 21 billion parameters, 3.6 billion active, and runs in 16 GB of memory⁠1openai/gpt-oss-120bOpenAI (Hugging Face). Both are text only. vLLM serves them as an OpenAI-compatible endpoint, which lets existing tools call them without changes⁠12openai/gpt-oss-120bOpenAI (Hugging Face)Online serving (OpenAI-compatible server)vLLM.

OpenAI documents fine-tuning the 120b model on a single H100 node and the 20b model on consumer hardware⁠1openai/gpt-oss-120bOpenAI (Hugging Face). The licence is Apache 2.0, accompanied by a short usage policy that asks users to comply with applicable law⁠3gpt-oss usage policyOpenAI (Hugging Face). Vendor-reported, gpt-oss-120b scores 80.81 on GPQA Diamond⁠1openai/gpt-oss-120bOpenAI (Hugging Face). Per-token endpoints are cheap: Together AI lists gpt-oss-120B at $0.15 per million input tokens and $0.60 per million output tokens⁠4Pricing: serverless inference, dedicated inference and GPU clustersTogether AI.

Strengths and limits

Strengths

• Apache 2.0 with a short usage policy.
• OpenAI documents fine-tuning the 120b on a single H100 node and the 20b on consumer hardware.
• Served as an OpenAI-compatible endpoint by vLLM.

Limits

• Text only.
• No new release since August 2025.
• Weights only: no training data.

Choose another option if: You need images or a newer generation (Gemma 4 or Qwen3.8-27B).

Pricing

Weights are free. Together AI lists gpt-oss-120B at $0.15 per million input tokens and $0.60 per million output tokens.

Questions to ask before choosing gpt-oss

• Is a text-only model enough for our work?
• Does a 2025 model still meet our quality bar against newer families of a similar size?
• Will we fine-tune the 20b model on our own hardware first, before committing to the 120b?
• Do we want one GPU per model, or a larger model behind a router for the hardest tasks?

See how gpt-oss compares with the other options in the main guide.

Sources

1. openai/gpt-oss-120b. OpenAI (Hugging Face). Accessed 9 Nov 2026.
2. Online serving (OpenAI-compatible server). vLLM. Accessed 9 Nov 2026.
3. gpt-oss usage policy. OpenAI (Hugging Face). Accessed 9 Nov 2026.
4. Pricing: serverless inference, dedicated inference and GPU clusters. Together AI. Accessed 9 Nov 2026.

Further reading

• Buyer's guide · Updated quarterlyThe best open-source AI in 2027, for companies that want to own itAn analysis of the open-weight model families a company can run today and the layer that turns one into its own, compared on seven published criteria, with sourced licences, costs and the established path from rented AI to owned AI.
• Practical guideHow to train your own AI model on company dataMost of the work in training a company model happens before any training starts. A step-by-step account of the data work, the choice of base model, the training methods and the rules that apply in Europe.
• Case analysisGoing headless like Revolut: what replacing Claude Code with an open model tookRevolut now runs most of its AI coding on an open-weight model it serves itself, for about a tenth of its former bill. A reconstruction of what it built, what the numbers mean for other companies, and the step that comes after.

Also in: Deutsch · Français · Español · Italiano · Nederlands · Polski