Claude Sonnet 3.5 → Claude Sonnet 4.5
In-place upgrade within the Claude Sonnet line.
A drop-in upgrade for almost every team. 4.5 is more reliable on long-context and parallel tool use; cost is unchanged. Audit any prompts that relied on 3.5's specific quirks; plan the swap as a one-line registry change.
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.
Capability gaps to plan for
- · Run your full eval suite against the new model in a Switchy Space before flipping production.
Side-by-side
See the full comparison: claude-sonnet-3-5 vs Anthropic: Claude Sonnet 4.5.