OpenAI's open-weights model, released under Apache 2.0. Running in the Berges fast-response layer.
gpt-oss is OpenAI's open-weights release. The gpt-oss-120b model is a balanced, instruction-tuned mixture-of-experts model that performs competitively on many tasks despite being open.
Berges AI runs gpt-oss in the fast-response layer, where it handles the bulk of everyday questions quickly and cheaply. When something needs more depth, the cascade escalates to one of the larger thinking-layer models.
It's published under the Apache 2.0 license, the most permissive of the major open-source licenses. If you want to host your own, you can.
OpenAI has years of experience tuning models to follow instructions cleanly. That shows in gpt-oss: it stays on task and respects formatting requests.
Not the best at any single thing, but solidly above the floor across most everyday tasks. A reliable default for the 90% of questions.
The most permissive major open-source license. You can use, modify, and redistribute the weights commercially with no copyleft obligations.
No. gpt-oss is OpenAI's open-weights model, released separately from the closed ChatGPT models (GPT-4o, GPT-4.1, etc.). It's competitive on many tasks but it's a different model.
gpt-oss-120b is well-tuned and runs efficiently. It's a strong default for the 90% of questions that don't need a larger reasoning model.
Yes. It's Apache 2.0, so you can download the weights and host them. Berges AI is one place to chat with it without that infrastructure overhead, with encryption at rest as a bonus.