Migrating: Sonnet 4.5 to Sonnet 5
Claude Sonnet 5 has itself been superseded by Claude Sonnet 5.5. If you are migrating now, check whether you can go straight there.
What changes
| From | To | |
|---|---|---|
| Model | Sonnet 4.5 | Sonnet 5 |
| Input / MTok | $3 | $2 |
| Output / MTok | $15 | $10 |
| Cache read / MTok | $0.30 | $0.20 |
| Min cacheable prefix | 1,024 | 1,024 |
| Context window | 200K | 1M |
| Effort levels | none | low, medium, high, xhigh, max |
| Effort default | — | high |
| Thinking | Fixed budget_tokens, off unless requested | On by default — can be disabled |
| Forced tool_choice | accepted | accepted |
Migrating from a generation with no effort control at all
This is a two-generation jump in effort-parameter terms: the source model predates the effort
parameter entirely, relying only on the older thinking-budget mechanism, while the destination
carries the full current range including its newest, highest tier. There's no existing effort
setting to carry forward here — the destination's default is being adopted from a blank slate
rather than migrated from a prior deliberate choice.
Why this deserves more testing than the adjacent-generation move
Because two generations of change are compressed into one migration, and because there's no prior effort setting to anchor a starting point against, this move needs more deliberate testing than a single-generation upgrade would — start from the recommended default effort level and validate against your actual task distribution, rather than assuming any specific starting point transfers cleanly from a model that never had this concept at all.
Treat this as a full migration, not a version bump
Given how much separates these two generations, this move deserves the same scrutiny you'd give any significant dependency upgrade — full re-testing against real traffic, not a quick smoke test against a couple of example prompts assumed to represent the whole workload. Start your effort-level choice from the recommended default, validate output quality and cost together against a representative sample, and only then consider whether the newest, highest tier is worth testing for your hardest cases specifically.
Where this move now leads
Sonnet 5 has been superseded by Sonnet 5.5 at the same prices. If you are starting this migration now, the effort sweep you run is better spent on Sonnet 5.5, whose effort levels are recalibrated — a level name there does not mean what it meant on Sonnet 5.
Related
See migrating off budget_tokens to the effort parameter for the mechanics of adopting effort-based control for the first time, and migrating: Sonnet 5 to Sonnet 5.5 for the last step.
Verified 2026-10-28 against claude-api skill — shared/models.md, model-migration.md + SKILL.md model table and Anthropic's pricing and model-deprecation pages (canonical: https://platform.claude.com/docs/en/about-claude/models/overview, https://platform.claude.com/docs/en/about-claude/pricing, https://platform.claude.com/docs/en/about-claude/model-deprecations).Could not confirm: Context window and max output are not restated in our source for this model; re-check via the Models API before budgeting. Amazon Bedrock and Google Cloud (Google managed) set their own retirement dates, which we do not track.
Verified 2026-09-30 against claude-api skill — shared/models.md, model-migration.md + SKILL.md model table and Anthropic's pricing and model-deprecation pages (canonical: https://platform.claude.com/docs/en/about-claude/models/overview, https://platform.claude.com/docs/en/about-claude/pricing, https://platform.claude.com/docs/en/about-claude/model-deprecations).