NOUMENOpen app
NOUMEN

OpenAI · Open weights

GPT OSS 20B

Added Aug 5, 2025Open weights

Compare models

Choose up to 3 models to compare.0 of 3 chosen

Loading models…

Use in the app

An open-weight 21B parameter model released under the Apache 2.0 license. It uses a Mixture-of-Experts (MoE) architecture with 3.6B active parameters per forward pass, optimized for lower-latency inference and deployability on consumer or single-GPU hardware. The model is trained in OpenAI's Harmony response format and supports reasoning level configuration, fine-tuning, and agentic capabilities including function calling, tool use, and structured outputs.

  • Reasoning
  • Tools
Input
Text
Output
Text
Discovery tier
Open weights

Pricing

USD per 1M tokens
Input
$0.02
Output
$0.10

Reference prices as published by the model provider. API example

Specifications

Context window
131.1K
Max output
16.4K
Average output (7 days)
1.9K tokens
Longer than 91% of 468 models

Side by side

Compare models

Put GPT OSS 20B next to up to three other models. The comparison stays in the link, so you can share it.

Compare models

Choose up to 3 models to compare.0 of 3 chosen

Loading models…

TryMistral Small 4 119BSarvam 105BQwen3 Coder NextAll 3

Public evaluations

Benchmarks

Independent evaluation results, with how each score ranks among the models in the Noumen catalog.

Intelligence Index

9.0

Tested at high reasoning effort

Better than 16% of 326 scored models

3.8Catalog models by score55.1

Indexes

  • Coding Index

    20.7

    Composite of the coding evaluations

    Better than 14% of 168 scored models

  • Math Index

    89.3

    Composite of the math evaluations

    Better than 86% of 118 scored models

Agentic work

  • AutomationBench-AA

    0.2%

    Workflow automation, penalized for guardrail violations

    Better than 0% of 173 scored models

  • AutomationBench-AA tasks completed

    0.0%

    Workflows finished end to end without a guardrail violation

    Better than 0% of 173 scored models

  • GDPval-AAv2.1

    339 Elo

    Economically valuable occupational tasks, rated head to head

    Better than 5% of 187 scored models

Document reasoning

  • GDP.pdf

    2.0%

    Professional reasoning over PDF documents, all-pass rate

    Better than 8% of 171 scored models

  • AA-LCRv1.1

    34.7%

    Reasoning across long documents

    Better than 16% of 309 scored models

  • MLCR-AA

    1.1%

    Long-context reasoning over medical records

    Better than 2% of 121 scored models

Reasoning and knowledge

  • Humanity’s Last Exam

    11.0%

    Expert-level questions across academic fields

    Better than 35% of 323 scored models

  • GPQA Diamond

    68.8%

    Graduate-level science questions

    Better than 28% of 291 scored models

  • MMLU-Pro

    74.8%

    Multi-subject knowledge and reasoning

    Better than 30% of 132 scored models

  • IFBench

    65.1%

    Following precise output instructions

    Better than 64% of 215 scored models

Coding

  • Terminal-Benchv4.0

    0.0%

    Practical software engineering in a terminal

    Better than 0% of 171 scored models

  • SciCode

    38.9%

    Scientific computing in Python

    Better than 18% of 174 scored models

  • LiveCodeBench

    77.7%

    Recent competitive programming problems

    Better than 78% of 129 scored models

Math

  • AIME 2025

    89.3%

    Competition mathematics, 2025 problems

    Better than 86% of 118 scored models

Legacy evaluations

  • Terminal-Bench Hard

    10.6%

    Earlier terminal task suite

    Better than 22% of 212 scored models

  • τ²-Bench Telecom

    60.2%

    Tool use in a customer-support simulation

    Better than 47% of 214 scored models

  • AA-LCR (unversioned)

    34.7%

    Long-context reasoning, earlier grading

    Better than 16% of 309 scored models

Scores from Artificial Analysis, updated Oct 26, 2026. GPT OSS 20B was tested at high reasoning effort; compare scores measured at the same effort.Artificial Analysis

Live routing data

Performance and reliability

Recent averages across requests routed for this model. Your own speed depends on prompt length, reasoning effort and load.

Prompt cache hits
0.0%
Request success (24h)
99.6%
Reliable

Cache hits are the share of input tokens recently served from a prompt cache across all requests. They are not a forecast of your own hit rate.

Hourly request success · last 72 hoursLowest hour 77.7%
  • Healthy (98% or more)
  • Degraded (90–98%)
  • Down (under 90%)
  • No requests

Developers

Use it through the API

Noumen’s API is OpenAI-compatible: any Chat Completions client works with a base URL and a key.

Read the quickstart
curl https://www.noumen.xyz/api/v1/chat/completions \
  -H "Authorization: Bearer $NOUMEN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/gpt-oss-20b",
    "messages": [{ "role": "user", "content": "Explain this stack trace." }]
  }'
import os
from openai import OpenAI

client = OpenAI(base_url="https://www.noumen.xyz/api/v1", api_key=os.environ["NOUMEN_API_KEY"])
reply = client.chat.completions.create(
    model="openai/gpt-oss-20b",
    messages=[{"role": "user", "content": "Explain this stack trace."}],
)
print(reply.choices[0].message.content)

OpenAI

Other OpenAI models, starting with the same family.