GPT OSS 20B (Groq)

GPT OSS 20B is OpenAI's compact open-weight reasoning model, released on August 5, 2025, alongside the larger gpt-oss-120b as part of the gpt-oss family. Both models are published under the Apache 2.0 license, marking OpenAI's first open-weight language model release since GPT-2 and enabling developers to download, fine-tune, and run the weights on their own hardware. The 20B variant is specifically sized for efficiency, designed to run on edge devices with roughly 16 GB of memory, which makes it well suited to on-device inference, local prototyping, and scenarios where sending data to a remote API is not desirable.

According to OpenAI, gpt-oss-20b was trained with a mix of reinforcement learning and techniques informed by OpenAI's most advanced internal systems, including o3, giving it a reasoning-oriented post-training approach rather than a purely supervised setup. OpenAI reports that the model delivers results comparable to o3-mini on common benchmarks while remaining small enough for consumer hardware, and both gpt-oss variants are highlighted for strong tool use, few-shot function calling, chain-of-thought reasoning, and competitive performance on agentic evaluations such as Tau-Bench and HealthBench, where they are said to even outperform some proprietary models. The 20B size is therefore aimed at developers who want frontier-style reasoning behavior and agentic workflow support without the infrastructure cost of a larger model.