OpenAI’s U-Turn: Sam Altman Admits Being ‘On Wrong Side of History’ About Open Source

OpenAI has reversed course. The company released its first open-source models since GPT-2 in 2019. This marks a major shift in strategy for one of the world’s most powerful AI companies.

Sam Altman made the announcement after OpenAI raised $40 billion at a $300 billion valuation. The company released two open-weight models: gpt-oss-120b and gpt-oss-20b. Both are available under the Apache 2.0 license. That means developers can use, modify, and build on them freely.

OpenAI wasn’t always closed. The company started in 2015 as an open-source, non-profit organization. Its name literally included the word “Open.” Over time, it shifted to keeping its models private. Now it’s walking that back.

OpenAI once stood for openness. Now, after years of going closed, it’s finally living up to its name again.

Competition played a big role in this change. DeepSeek, a Chinese AI startup, shook the industry in January. It released powerful open-weight models that rivaled closed systems. Meta also pushed open-source AI forward with its own model releases. These moves put pressure on OpenAI to rethink its approach.

The new models aren’t just free — they’re also strong performers. OpenAI says gpt-oss-120b and gpt-oss-20b beat similarly sized open-source competitors on reasoning tasks. They’re also built for efficient use on consumer hardware. That makes them cheaper to run.

Training used reinforcement learning from advanced internal models, including o3.

Pricing for the hosted versions is low. The 120 billion parameter model costs 15 cents per million input tokens. Output tokens run 60 cents per million. That’s affordable for startups and smaller teams.

The open-source move also fits a broader goal. OpenAI says it wants individuals, businesses, and governments to run and customize AI on their own. Users get access to the model weights and code. They can adjust the models for their needs. Researchers at Dartmouth College have shown that AI-powered applications can extend access to underserved populations, reinforcing why open models matter beyond just developer convenience.

Industry insiders say competition wasn’t the only reason for the switch. But the timing lines up with rising pressure from open-source alternatives. Open-source models also allow developers to customize for specific use cases, reducing their dependence on proprietary systems like OpenAI’s. ChatGPT recently reached 500 million weekly users, reflecting how quickly OpenAI’s influence has grown alongside these strategic shifts.

OpenAI’s shift shows how fast the AI market’s moving. What was once a firm strategy can change quickly when rivals close the gap.