GPT-4 → Claude Sonnet 4.5
Switching from OpenAI GPT-4 to Anthropic Claude Sonnet 4.5.
The most-asked migration on Switchy. Claude follows instructions more literally and is stricter about role-play; expect to lighten your system prompts. Tool-use shape is similar but not identical — review parallel tool calls if you use them heavily.
Prompt translation tips
- · Start with your existing system prompt unchanged and run a 20-prompt evaluation set side-by-side; only change what fails.
- · Watch for instruction-following drift — different labs follow instructions at different temperatures (Anthropic models are stricter).
- · Tool-call shapes vary — review parallel tool calls, JSON-only output requirements, and any structured-output hints you depend on.
- · If you use vision, retest your prompt set with the same images — different models excel at different layouts.
Pricing math
Moving from OpenAI: GPT-4.1 to Anthropic: Claude Sonnet 4.5:
- Input tokens: +$1.00/Mtok (more expensive).
- Output tokens: +$7.00/Mtok (more expensive).
Capability gaps to plan for
- · Context window shrinks from 1048K to 1000K — plan for chunked retrieval if you load whole documents.
- · Run your full eval suite against the new model in a Switchy Space before flipping production.
Side-by-side
See the full comparison: OpenAI: GPT-4.1 vs Anthropic: Claude Sonnet 4.5.