OpenAI Joins an Open-Weight Coalition, Reversing a Decade of Closed-Model Strategy

On July 24, 2026, OpenAI added its name to a public letter supporting open-weight AI models, joining a coalition that has now grown to 35 organizations. The original group of 25 signatories was led by Microsoft, NVIDIA, IBM, and Meta, and the latest additions include GitHub, Cohere, Cisco, Palo Alto Networks, and the open-source lab Nous Research. Together they argue that letting anyone download, inspect, and run model weights on their own hardware is essential for the next stage of the industry. The letter matters because OpenAI has spent almost a decade keeping its frontier models closed-weight. The company's earlier models, including several in the GPT family, were only available through paid APIs. By signing, OpenAI is now pledging to advocate for open-weight development and to publish weights for at least some of its capable systems, even as it continues to ship closed products. Supporters say open weights let universities, startups, and governments build sovereign AI infrastructure without depending on a single vendor. Critics inside the company warn that releasing powerful weights could accelerate misuse, from cyber attacks to large-scale disinformation. The letter tries to answer those fears with a shared set of safety and reporting commitments that every signatory agrees to follow. For the wider industry, the shift closes a long-running argument about whether frontier research must stay locked behind APIs. Closed-weight labs now face clear market and policy pressure to explain why a given model cannot be released, and open-weight projects are likely to see stronger institutional backing through 2027.