GPT OSS

OpenAI's first open-weight models since GPT-2, Apache 2.0 licensed

OpenAI released gpt-oss-120b and gpt-oss-20b — two open-weight language models under the Apache 2.0 license, the company's first open-weight release since GPT-2. The 120B model achieves near-parity with o4-mini on core reasoning benchmarks while running on a single 80GB GPU. The 20B model matches o3-mini on common benchmarks and runs on devices with just 16GB of memory. Both models were trained using reinforcement learning informed by OpenAI's frontier models including o3, and they demonstrate strong tool-use capabilities critical for agentic applications. They're available on Hugging Face, Azure AI Foundry, and can be run locally on consumer hardware via NVIDIA and AMD optimization paths. The release marks a strategic shift for OpenAI, which had ceded the open-weight market to Meta's Llama series and Alibaba's Qwen family. GPT OSS challenges Meta's open-source dominance directly, and the Apache 2.0 license (not the restrictive license used by some Llama releases) makes it genuinely deployable in commercial products without royalties.

Panel Reviews

Ship

“Apache 2.0 and near-o4-mini performance at 120B is the combination I've been waiting for. This is the first open-weight model I can actually build a commercial product on without legal uncertainty.”

Skip

“OpenAI is releasing open weights now that their closed models are multiple generations ahead. This is strategic positioning, not altruism. Also, Llama 4 and Qwen 3.6 are already in market — GPT OSS needs a real edge to win mindshare.”

Skip

“The frontier open-weight race just got a new major player. When OpenAI, Google, Meta, and Alibaba are all competing on open weights, the floor for local AI capabilities rises dramatically for everyone.”

Ship

“The 20B model running on 16GB of RAM is the one I'm excited about. Fine-tunable, commercially usable, and good enough for most production content tasks. This is a big deal for indie devs.”