Price, first
Sonnet 5 is $2 input / $10 output per million tokens. GPT-5.6 Sol is $5 / $30. Opus 5 is $15 / $75. For a busy month of agent traces, Sonnet 5 is the bill you can keep. Sol is the OpenAI flagship you pick when the rest of the stack is already there.
Rankings we actually mirror
On the August 2026 LMArena Agent board we publish: Opus 5 is #1, GPT-5.6 Sol is #4, Sonnet 5 is #7. That is why “Claude vs GPT” is the wrong question if you mix Opus and Sonnet. Name the SKU.
Tooling
Claude Code is the first-party CLI. GPT-5.6 Sol shows up across Codex-style and third-party agents. Cursor can drive both. If your team already standardizes on Anthropic, stay there unless Sol is required by a vendor integration.