claude-sonnet-5 — pricing, context window and limits
The previous Sonnet, still served at Sonnet 5.5's prices — its introductory launch price became permanent. Its tokenizer counts roughly 30% more tokens than Sonnet 4.6's for the same text, and adaptive thinking is on by default.
Price per million tokens
| Input | $2 |
|---|---|
| Output | $10 |
| Cache write, 5-minute TTL | $2.50 |
| Cache write, 1-hour TTL | $4 |
| Cache read | $0.20 (0.1x input) |
| Batch input / output | $1 / $5 |
Limits and behaviour
| Status | Current |
|---|---|
| Context window | 1M tokens |
| Max output | 128K tokens |
| Min cacheable prefix | 1,024 tokens |
| Effort levels | low, medium, high, xhigh, max |
| Effort default | high |
| Thinking | On by default — can be disabled |
| Forced tool_choice | Accepted |
| Platforms | Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry |
| Released | 30 Jun 2026 |
Lifecycle
| API model name | claude-sonnet-5 |
|---|---|
| Retires no sooner than | 30 Jun 2027 |
| Successor | Claude Sonnet 5.5 |
Its launch price became its standard price
Sonnet 5 launched at a price Anthropic described as introductory, with an increase scheduled for a later date. That increase never happened: Anthropic's pricing page now states the launch price is the standard price. Anything budgeted against the higher figure — including this site's own calculators, for a short stretch — overstated Sonnet 5's cost. See Sonnet 5 kept its launch price for how that played out.
The previous Sonnet, at the same price as the new one
Sonnet 5.5 replaced Sonnet 5 in the Sonnet line at identical per-token prices and with the same
tokenizer, so the move costs nothing on the rate card. What Sonnet 5 keeps is behaviour: it still
accepts thinking: disabled and forced tool choice, both of which Sonnet 5.5 rejects with a 400.
Its minimum cacheable prefix is also higher than Sonnet 5.5's.
The two traps it introduced over Sonnet 4.6
Its tokenizer counts noticeably more tokens than Sonnet 4.6's for identical text, so a per-token
price cut is not automatically a bill cut. And thinking runs by default, which it didn't on the
prior generation, so max_tokens has to cover thinking as well as the answer. See
what actually changed from Sonnet 4.6 to Sonnet 5
for both in full.
Related
See migrating: Sonnet 5 to Sonnet 5.5 for the breaking changes, and claude-sonnet-5-5 for the successor.
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).
Verified 2026-10-28 against Anthropic's model-deprecations status table and pricing page (canonical: https://platform.claude.com/docs/en/about-claude/model-deprecations, https://platform.claude.com/docs/en/about-claude/pricing).Could not confirm: These dates apply to the Claude API, Claude Platform on AWS, Claude Models on Google Cloud (Anthropic managed) and Microsoft Foundry. Amazon Bedrock and Claude Models on Google Cloud (Google managed) keep their own retirement schedules, which we do not track.