OpenAI’s latest language model, GPT-5, is now available to be used in Cursor, a tool designed for coding tasks. The team at Cursor has started using GPT-5 internally and shares their initial impressions of the model.
According to five engineers who tested GPT-5, it excels in complex tasks, providing solutions that other models couldn’t figure out. For instance, they used GPT-5 with Background Agents for independent tasks and parallel foreground agents when switching between different tasks.
GPT-5 has also proven itself to be highly steerable, allowing users to guide the model towards specific outcomes. The team discovered this capability while using it for writing code, where they could steer the model to produce concise responses or avoid follow-up questions.
One notable example was a test where GPT-5 successfully generated code for a backend API endpoint and corresponding frontend React component in a single shot. This showcased its ability to regenerate protobufs to fix type errors.
The team describes GPT-5 as intelligent, steerable, and well-suited for coding tasks. During the launch week, users can access GPT-5 with free credits through OpenAI’s partnership with Cursor.
Source: https://cursor.com/blog/gpt-5