TL;DR

• Claude Opus 4 leads in raw performance and prompt adherence.
• It understands user intentions better, reminiscent of 3.6 Sonnet.
• High taste. The generated outputs are tasteful. Retains the Opus 3 personality to an extent.
• Though unrelated to code, the model feels nice, and I never enjoyed talking to Gemini and o3.
• Gemini 2.5 is more affordable in pricing and takes fewer API credits than Opus.
• One million context length is undefeatable for large codebase understanding.
• Opus is the slowest in time to first token. You have to be patient with the thinking mode.