Everyday questions and drafts
For the steady stream of emails, explanations and quick rewrites, GPT-6.1 Sol is fast, confident and broadly capable. Opus 5.5 is more than you need here; Claude's own everyday model, Sonnet 5.5, is the fairer match. A typical Sol reply costs 16 tokens on LazySusan against 31 tokens for Opus 5.5.
edge: GPT-6
Long documents and careful analysis
Opus 5.5 is at its best reading something long and finding what's inconsistent, missing or quietly important. It tends to quote the passage it relies on and to say when the text doesn't support a conclusion. GPT-6 Astra handles long material well too, but Claude's habit of showing its working makes it easier to check.
edge: Claude Opus 5.5
Code
Both are strong. Opus 5.5 is particularly good when it can see several files at once and has to trace a bug across them; GPT-6 Astra is excellent at hard algorithmic problems and at writing a lot of correct code quickly. Many developers keep both and use the second as a reviewer for the first.
edge: Tie
Writing with a voice
If the output has to sound like a person, especially a specific person, Claude usually needs fewer rounds of editing. GPT-6 writes clean, well-organised prose and is better at punchy, formatted output such as lists, tables and slide outlines.
edge: Claude Opus 5.5
How much context you can send
On LazySusan, GPT-6 models accept prompts up to 400K tokens and Claude models up to 1M. Very large prompts cost more on both, and OpenAI bills requests over 272K input tokens at a higher rate.
edge: Claude Opus 5.5
Cost per message
GPT-6 Astra and Claude Fable 5.1 are the two premium options, at about 77 tokens for a typical reply each. Opus 5.5 costs 31 tokens and Sol 16 tokens. For quick jobs, GPT-6 Luna and Claude Haiku 5.5 are each about 1 token.
edge: GPT-6