Model by OpenAI

gpt-oss-20b

gpt-oss-20b is OpenAI's smaller open-weights model, suited to running a reasoning model on your own machine or paying very little through an API.

Checked 2026-11-05

gpt-oss-20b is OpenAI's smaller open-weights model, suited to running a reasoning model on your own machine or paying very little through an API. The Hugging Face card lists 21B parameters in total with 3.6B active, says it runs within 16GB of memory thanks to MXFP4 quantisation of the expert weights, and offers low, medium and high reasoning levels set in the system prompt. The licence is Apache 2.0, a permissive licence with no copyleft terms on the card. OpenRouter listed it on 5 August 2025 with a context of 131,072 tokens, text only, at $0.03 per million input tokens and $0.13 per million output tokens on its cheapest route, as listed on 5 November 2026. The card gives no context length, so that figure comes from OpenRouter alone. It is an August 2025 model, and gpt-oss-120b is the larger sibling. We have not run it ourselves. We would use it for local experiments, cheap bulk tasks and offline work, not for the hardest coding jobs. Figures come from the Hugging Face card and OpenRouter's list.

Mentioned here

  • gpt-oss-120b

    gpt-oss-120b is OpenAI's open-weights model under Apache 2.0, a standard permissive licence, with a context of about 131,072 tokens.

    • Open weights

    OpenAI, 131K context, $0.15 in / 1M

  • Gemma 4 26B A4B

    Gemma 4 26B A4B is Google DeepMind's mid-sized open model, and it suits people who want image and video understanding with low active compute under Apache 2.0.

    • Open weights

    Google, 262K context, $0.03 in / 1M

  • Phi 4

    Phi 4 is a small open model from Microsoft, released 2025-01-10 under the MIT licence, listed on OpenRouter at $0.07 per million input tokens and $0.14 output.

    • Open weights

    Microsoft, 16K context, $0.07 in / 1M

  • Ollama

    Ollama runs open models on your own machine and also offers hosted cloud models.

    • Self-hosted runtime
    • OpenAI compatible
    • Self-hostable

  • OpenAI API

    OpenAI's API bills per token.

    • Model / inference API
    • OpenAI compatible

Questions

What is the context window of gpt-oss-20b?

131,072 tokens, with up to 32,768 tokens of output.

How much does gpt-oss-20b cost?

$0.03 per million input tokens and $0.13 per million output tokens, checked 5 November 2026.

Can I run gpt-oss-20b on my own hardware?

Yes, the weights are published under the Apache 2.0 licence. They are on huggingface.co.

Who makes gpt-oss-20b?

OpenAI makes it, and released it on 5 August 2025.

Reviews

Loading reviews…

Write a review

Your rating

Up to 1000 characters. Stars alone are enough.

Where you can run it

Compare hosts

  • OpenRouter

    OpenRouter routes more than 600 models through one API key.

    • Model gateway (router)
    • OpenAI compatible

Similar

  • GPT-4.1

    GPT-4.1 is OpenAI's April 2025 non-reasoning model, built for instruction following and long-context code work, and it is now deprecated.

    • Closed

    OpenAI, 1M context, $2 in / 1M

  • o3

    o3 is OpenAI's April 2025 reasoning model, and it is now deprecated, so it suits only code that still calls it.

    • Closed

    OpenAI, 200K context, $2 in / 1M

  • GPT-5.4 Nano

    GPT-5.4 Nano was OpenAI's cheapest GPT-5.4 model, built for speed-critical, high-volume jobs, and it is now deprecated.

    • Closed

    OpenAI, 400K context, $0.20 in / 1M

  • GPT-5.4 Mini

    GPT-5.4 Mini is OpenAI's smaller GPT-5.4 model, aimed at high-throughput coding and reasoning work where the full model costs too much.

    • Closed

    OpenAI, 400K context, $0.75 in / 1M

  • GPT-5.4

    GPT-5.4 is an earlier OpenAI frontier model, and it suits teams already on the OpenAI API who need a very long context for coding and document work.

    • Closed

    OpenAI, 1.1M context, $2.50 in / 1M

  • GPT-5.3-Codex

    GPT-5.3-Codex is OpenAI's agentic coding model from early 2026, and it now suits only teams that must keep an existing pipeline running.

    • Closed
    • Coding

    OpenAI, 400K context, $1.75 in / 1M