gpt-oss-20b
gpt-oss-20b is the smaller of OpenAI's two August 2025 open-weight models: a 32-expert mixture-of-experts with 21B total and 3.6B active parameters. OpenAI says it runs within 16 GB of memory, so it can run on many consumer GPUs and edge devices. It keeps the 128K context and adjustable reasoning effort of the larger model.
Specifications
- Context window
- 128Kabout 98,000 words, or 200 pages
- Max output
- Not published
- Input price
- $0.075per 1M tokens
- Output price
- $0.30per 1M tokens
- Cached input
- Not published
- Weights
- OpenApache-2.0
- Parameters
- 21B total, 3.6B activeMoE
- Input types
- Text
- Output types
- Text
- Reasoning mode
- Yes
- Tool calling
- Yes
- Released
- Aug 5, 2025Current
Groq price for openai/gpt-oss-20b (Groq allows up to 65,536 completion tokens; OpenAI states no separate output limit).
What it costs in practice
| Workload | Monthly cost at list price |
|---|---|
| Support chatbot30,000 conversations a month, 2,000 input and 400 output tokens each | $8.10 |
| Document summaries2,000 documents a month, 12,000 input and 800 output tokens each | $2.28 |
| Coding agent1,000 tasks a month, 60,000 input and 6,000 output tokens each | $6.30 |
Standard rates, no caching or batch discounts. Run your own numbers.
Strengths
- Runs within 16 GB of memory
- Adjustable reasoning effort and tool-use formats
- Apache 2.0 license
Trade-offs to weigh
- Text-only input
- Limited world knowledge for its size
Where it fits
OpenAI positions gpt-oss-20b for:
Where you can use it
- Hugging Face
- Ollama
- Groq
- OpenRouter
- AWS
- Azure
Run it yourself
Released under Apache-2.0. Approximate memory needed just to load the weights:
| Precision | Memory for weights |
|---|---|
| 16-bit (BF16/FP16) | 42 GB |
| 8-bit (FP8/INT8) | 21 GB |
| 4-bit (Q4_K_M, about 4.85 bits) | 13 GB |
This is a mixture-of-experts model: all 21B total parameters must sit in memory, but only about 3.6B are used per token, so it runs faster than a dense model of the same size.
Add memory for the KV cache (it grows with context length and concurrent users) and 1-2 GB of runtime overhead. Size it precisely in the VRAM calculator.
Alternatives to consider
| Model | Tier | Context | Input / output per 1M | |
|---|---|---|---|---|
| Gemma 4 26B A4BGoogle | Compact | 256K | $0.07 / $0.34 | Compare |
| Ministral 3 8BMistral AI | Compact | 256K | $0.15 / $0.15 | Compare |
| Granite 4.2 8BIBM | Compact | 128K | $0.06 / $0.25 | Compare |
| Ministral 3 3BMistral AI | Compact | 256K | $0.10 / $0.10 | Compare |
Frequently asked questions
How much does gpt-oss-20b cost?
gpt-oss-20b lists at $0.075 per million input tokens and $0.30 per million output tokens (checked October 10, 2026). Groq price for openai/gpt-oss-20b (Groq allows up to 65,536 completion tokens; OpenAI states no separate output limit). Confirm current rates on the provider's pricing page before you commit.
What is the context window of gpt-oss-20b?
gpt-oss-20b accepts up to 131,072 tokens of context (about 98,000 words, or 200 pages). Long prompts cost more and can be slower, so most apps send only what the task needs.
Is gpt-oss-20b open source?
gpt-oss-20b is an open-weight model released under the Apache-2.0 license. You can download the weights from Hugging Face (openai/gpt-oss-20b) and run or fine-tune them yourself, within the license terms. Open weights isn't the same as fully open source: training data and code may not be published.
Can I run gpt-oss-20b locally?
Yes, if you have enough memory. The weights alone need about 13 GB at 4-bit quantization and about 42 GB at 16-bit, plus room for the KV cache. Use the VRAM calculator on LLModel.com to check a specific GPU or Mac.
What is gpt-oss-20b best for?
OpenAI positions gpt-oss-20b for on-device, low latency, self-hosting, high volume. It accepts text as input, has a reasoning (thinking) mode and supports tool calling. Test it on a sample of your own tasks before you choose it for production.
Sources
- openai.com/index/introducing-gpt-oss
- console.groq.com/docs/models
- huggingface.co/openai/gpt-oss-120b
- huggingface.co/openai/gpt-oss-20b/raw/main/config.json
- ollama.com/library
Facts checked against these pages on October 10, 2026. Spot an error? Tell us.
Still choosing a model?
Send us your use case and constraints. We'll come back with a shortlist and an evaluation plan you can run on your own data.
- Free and vendor-neutral: no model provider pays for placement.
- Your details go only to partners you agree to.