GPT OSS 20B

GPT OSS 20B is part of OpenAI's gpt-oss family of open-weight large language models, released as Apache 2.0 licensed weights that can be downloaded, run locally, and fine-tuned on user-controlled hardware. The family marked OpenAI's first public release of internal model parameters since GPT-2, framing the gpt-oss lineup as an alternative to API-locked proprietary systems. Community coverage of the announcement appeared on the same date as the model's availability, positioning the 20B variant alongside its larger sibling as a flexible option for developers who want transparency and on-device control.

Beyond the model weights themselves, an MXFP4-quantized GGUF build of GPT-OSS-20B is distributed through community channels for local inference, with the Qualcomm AI Hub listing a roughly 11.27 GB artifact packaged for Snapdragon-class devices. The same deployment listing links the model to arXiv paper 2508.10925 as the associated research reference and points to a Hugging Face repository at qualcomm/GPT-OSS-20B where the quantized files are hosted. A llama.cpp-based runtime path is documented for Windows, Android, and Linux, with reported prefill and decode throughput on Snapdragon X2 Elite hardware, illustrating the model's fit for edge and local experimentation rather than purely cloud-hosted serving.

Groqopenai/gpt-oss-20bgpt-oss

Open provider page Provider docs

Quick Info

Provider: Groq
Model key: openai/gpt-oss-20b
Release date: Aug 5, 2025
Last updated: Sep 25, 2025
Input modalities
Output modalities
Capabilities

Cost

Input token cost: $0.075
Output token cost: $0.30

Limits

Output tokens: 65,536 tokens
Context window: 131,072 tokens

Transparent token rates

Compare GPT OSS 20B pricing

Rates are shown per one million tokens. Combined means one million input plus one million output tokens.

Browse this family

Latest news about GPT OSS 20B

Groq

Official source

I am using strict mode and the model is failing to genereate valid json. Is this expected? GroqException - {\“error\”:{\“message\”:\“Failed to generate JSON. Please adjust your prompt. See ‘failed_generation’ for more d…

Opper

Coverage

OpenAI released gpt-oss-20b on August 5, 2025 as an open-weight model with 21 billion total parameters, of which 3.6 billion activate per token, built around a Mixture-of-Experts Transformer architecture that combines dense attention with a local banded sparse attention mechanism and supports a 128,000-token context le According to the model's technical report, its performance is comparable to o3-mini, with strong results on benchmarks such as AIME and HealthBench, and in local deployment tests a quantized GPT-OSS 20B weighing 11.95 GB loaded in 8.3 seconds. The encyclopedia entry also documents distribution milestones: on August 6,

Groq

CoverageBenchmark

A new arXiv preprint (2607.13248v1, dated 14 Jul 2026) introduces GSM-Plus-BN, a perturbation-based benchmark of 9,000 Bengali math samples (1,000 seed questions plus 8,000 perturbed variants) derived from GSM-Plus and verified by human translators. The authors evaluate six open-source LLMs — Qwen3-32B, Llama-3.1-8B-In On this benchmark, GPT-OSS-20B achieves the highest seed-question accuracy of 96.08% under Standard Prompting, edging out the larger models tested. However, bigger models such as Llama-3.3-70B-Versatile and GPT-OSS-120B demonstrate superior robustness across the perturbation types, and CoT prompting substantially impro