OpenAI o3 vs Gemini 2.5 Pro: my actual use cases, not benchmarks

Benchmarks are mostly useless for determining which model to pick for your specific work. Here's what I found after six weeks alternating between them:

**Code generation (TypeScript, Python)**: o3 writes cleaner function signatures and handles edge cases in error handling better. 2.5 Pro is faster and cheaper for the 'write me a quick script' use case.

**Reasoning over long documents**: 2.5 Pro's 1M context window is genuinely useful. o3 is better at the actual reasoning once the document is short enough.

**Math**: o3, not close.

**Creative writing**: I find 2.5 Pro less stilted but this is almost entirely personal preference.

Cost note: o3 is ~4x more expensive per token than 2.5 Pro. For most use cases that aren't PhD-level reasoning, 2.5 Pro is the better value. I'd only reach for o3 when the task specifically benefits from extended thinking.