My Week with GPT-5

15 August 2025

Last week, OpenAI released the much-anticipated GPT-5. The reviews have been mixed, so much so that the company restored the 4o chat model because a vocal group of users were unhappy it went away.

My primary workflow lately includes Cursor, so I was able to take advantage of the free week of GPT-5. I have to say, for code use, it was impressive. There were several non-trivial updates I needed to make to a large iOS app feature, and it did some very heavy lifting. I was surprised to get multiple working solutions on the first request for complex functionality involving Core Graphics.

Cost Comparison with Claude Opus

GPT-5 is significantly cheaper than Claude Opus, with comparable if not better capabilities:

Token pricing per 1 million tokens
Model Input cost Output cost
GPT-5 (Standard) $1.25 $10.00
Claude Opus (4.1) $15.00 $75.00

Different Models for Different Situations

I’ve become more discerning about what models to use in what scenarios. My coding workflow going forward, where AI is used, will be:

  • GPT-5 or Gemini 2.5 for initial buildout of large areas, updates to complex features.
  • Claude Sonnet 4 for day to day, smaller work.