New Models from OpenAI

GPT-OSS for Self-Deployment
https://openai.com/index/introducing-gpt-oss/
OpenAI, a company known for its powerful but mostly closed GPT models, recently introduced a new family of models with open weights called GPT-OSS (Open Source Software). These can be run locally or on your own infrastructure.

This is a significant step, as GPT-OSS is OpenAI's first "open" release since GPT-2. The models have an MoE architecture. They are designed for complex reasoning tasks, agentic tasks (Python code execution and web browsing), and general-purpose use.

The models are deployed by many providers.
https://openrouter.ai/openai/gpt-oss-120b
After the release of GPT-OSS, the developer community began testing their capabilities. And although, according to OpenAI's charts, these are very cool models, they show average results, especially the smaller 20b model, when it comes to code generation. Perhaps their fine-tuned versions will appear in the near future.

GPT-5 for Beautiful Interfaces
https://openai.com/index/introducing-gpt-5/
The new flagship model, a new standard for others. It can regulate its efforts and work with code for a long time in an agentic manner.
https://cursor.com/blog/gpt-5
At the presentation, one of the Cursor developers said that this is now the main model for their application and can be directly used in your work, not just for test projects.

https://www.youtube.com/watch?v=BUDmHYI6e3g

https://openai.com/index/introducing-gpt-5-for-developers/
It demonstrates significant improvements in front-end generation and understanding large repositories. It more often creates more beautiful and responsive websites, applications, and games. The model is currently available in almost all major AI programming products: Cursor, Windsurf, Vercel, JetBrains, Factory, Lovable, Gitlab, Augment Code, GitHub, Cognition.

It should become a better assistant for programmers and less annoying and reluctant to work.

#newllmmodel