gpt-oss-20b
OpenAI's gpt-oss-20b is an open-weight, 21-billion parameter model released under the Apache 2.0 license. It utilizes a Mixture-of-Experts architecture to balance performance with efficiency, supporting both reasoning tasks and tool calling.
gpt-oss-20b specs and pricing
- Provider: OpenAI
- Released: August 5, 2025
- Context window: 131K tokens
- Max output: 118K tokens
- Input: text
- Output: text
- Price: $0.018 input / $0.09 output per 1M tokens
- Reasoning: yes
- Tool calling: yes
What gpt-oss-20b is good at
- Complex reasoning tasks requiring logical step-by-step processing
- Integrating AI capabilities into custom software via tool calling
- High-volume text processing applications
To-dos to try with gpt-oss-20b
- Analyze a complex technical document and break down the logical steps required to implement the described system.
- Use tool calling to query a database and summarize the results based on specific criteria.
- Draft a detailed project plan that accounts for multiple dependencies and potential risks.
FAQ
- What is the architecture of gpt-oss-20b? The model uses a Mixture-of-Experts (MoE) architecture with 3.6 billion active parameters per forward pass.
- Does this model support external tools? Yes, the model is designed to support tool calling, allowing it to interact with external functions.
- How much does gpt-oss-20b cost? gpt-oss-20b costs $0.018 per million input tokens and $0.09 per million output tokens through OpenRouter. In todo.is a typical agent step costs about 0.06 credits ($1 = 100 credits).
- What is the context window of gpt-oss-20b? gpt-oss-20b can read up to 131K tokens at once and write up to 118K tokens in one reply.
- Can I use gpt-oss-20b without coding? Yes. In todo.is you pick gpt-oss-20b as your agent's model and write to-dos in plain words, on the web, the Android app, WhatsApp or Telegram. No API key needed: credits are included, or you can bring your own key.
JavaScript is required to use the todo.is app.