Which model for what. Claude, GPT, Gemini

The big three

Claude (Anthropic). Strong at code, writing, long analyses. Very good at following complex instructions. Preferred by developers.

GPT-5 (OpenAI), currently the GPT-5.5 generation. Broadest knowledge, strongest ecosystem (ChatGPT, plugins, GPT Store). Solid as a default, less steerable in nuance.

Gemini (Google). Huge context window, good multimodal support (video, audio, image). Knowledge integrated from Google Search.

When to use which

For code and architecture work: the current Claude Opus version. Best instruction-following (May 2026: Opus 4.8, changes with each release).

For general research and writing: GPT-5 or Claude, almost on par.

For very long documents (500+ pages): the current Gemini Pro version (2026: Gemini 3.5 or Gemini 3 Pro). Very large context windows are standard here.

For multimodal tasks (video, audio): Gemini.

For agentic workflows (long tool-use chains): Claude. The Sonnet/Opus model is the standard for agent-building.

Pricing under control

Not everything needs the most expensive model. Roughly:

• Simple question, short answer: Haiku, GPT-5-mini, Gemini Flash
• Medium complexity, requires thinking: Sonnet, GPT-5, Gemini Pro
• Critical analysis, architecture, code review: Opus, GPT-5.5, Gemini 3 Pro

Per token, the tiers differ by a factor of 10. "Opus everywhere" is throwing money away.

Practical tip

Don't lock in. Try the three on the same task. The output difference is often bigger than the price difference. And what was best last week may be different this week.