OpenAI has released two new open-weight language models, gpt-oss-120b and gpt-oss-20b, under the Apache 2.0 license.

The company said these models are optimized for reasoning tasks and can be deployed on consumer hardware, with gpt-oss-120b running on a single 80 GB GPU and gpt-oss-20b functioning on devices with 16 GB of memory.

OpenAI reported that both models perform competitively with its proprietary models on academic benchmarks in coding, mathematics, and health tasks.

The models support various agentic workflows, including instruction following and tool use, and allow developers to adjust reasoning effort for different latency needs.

OpenAI said that the models underwent safety evaluations, including adversarial testing, and were reviewed by external experts.

This marks the company’s first release of open-weight language models since GPT-2.

🔗 Source: OpenAI

🧠 Food for thought

1️⃣ OpenAI breaks six-year strategic pause on open-weight releases

OpenAI’s decision to release open-weight models marks its first such move since GPT-2 in 2019, representing a significant strategic shift after maintaining a closed approach for six years 1.

This extended pause reflected OpenAI’s previous strategy of keeping advanced models proprietary to maintain competitive advantage and address safety concerns.

The timing of this release suggests mounting competitive pressure, as the AI landscape has evolved dramatically since 2019 with multiple players now offering capable open-weight alternatives.

The gap also shows how OpenAI’s business model priorities have shifted, from the early research-focused organization that released GPT-2 to the commercially-driven company that kept GPT-3, GPT-4, and other advanced models closed.

2️⃣ Market bifurcation forces major AI companies to adopt dual strategies

OpenAI’s release reflects the broader strategic reality that AI companies now must navigate both proprietary and open-weight ecosystems simultaneously.

The emergence of powerful open-weight models from companies like Meta has reshaped the competitive landscape, forcing even closed-model leaders to reconsider their approach 2.