My impression of GPT-5

2025-08-11

Summary: A sourced summary of one user's experience with GPT-5, including expectations, early disappointment, coding use, routing issues, and instruction-following strength.

This sourced note describes the author's first impressions after the GPT-5 launch. He expected a major jump, felt the launch presentation was underwhelming, then found the model more impressive when tested on complex work and coding tasks.

The main praised qualities are instruction following, tool use, and long-context behavior. The post also argues that some negative first impressions came from product and routing issues rather than the model's core capability. This is one person's experience, but it is useful for understanding how model quality and user-interface choices can create very different impressions.

Free Basics version: when judging an AI tool, test it on real tasks and check whether settings, limits, or routing may be affecting the result.

Source: Daniil Sedov, Gusarich's thoughts, gusarich.com/blog/my-impression-of-gpt-5