claude-3-haiku-20240307 — pricing, context window and limits
Retired on the Claude API on 2026-04-20. Migrate to Claude Haiku 5.5, the current Haiku.
Price per million tokens
| Input | $0.25 |
|---|---|
| Output | $1.25 |
| Cache write, 5-minute TTL | $0.313 |
| Cache write, 1-hour TTL | $0.50 |
| Cache read | $0.025 (0.1x input) |
| Batch input / output | $0.125 / $0.625 |
Limits and behaviour
| Status | Retired on the Claude API |
|---|---|
| Context window | 200K tokens |
| Max output | 4,096 tokens |
| Min cacheable prefix | 2,048 tokens |
| Effort levels | not supported |
| Effort default | — |
| Thinking | Fixed budget_tokens, off unless requested |
| Forced tool_choice | Accepted |
| Platforms | none confirmed |
Lifecycle
| API model name | claude-3-haiku-20240307 |
|---|---|
| Deprecated | 19 Feb 2026 |
| Retired | 20 Apr 2026 |
| Successor | Claude Haiku 5.5 |
Retired on the Claude API
Haiku 3 has been retired on the Claude API, on the date the lifecycle table above shows. When this page was first written we could only say its published retirement date had passed without our confirming the id stopped resolving; Anthropic's lifecycle table now records the retirement as executed, so a request to it on the first-party API fails.
The oldest naming convention on this site
Unlike every other model covered here, this id carries an explicit date suffix rather than the short, generation-based name the rest of the lineup uses. This site never constructs or extends ids of that shape, and seeing one in your code is a reliable sign of how old the integration around it is.
What we could not confirm
Anthropic's current pricing and caching pages no longer list this model at all, so the figures above are the last ones we held, and we could not confirm whether any partner platform still serves it. They are excluded from every calculator on this site.
Where to go
Anthropic's deprecation notice named Haiku 4.5 as the replacement, and its migration guide now points Haiku workloads at Haiku 5.5, the current Haiku. Haiku 5.5 is cheaper per token than Haiku 4.5, but it is priced by prompt length and rejects several request shapes Haiku 3 accepted, so re-price the workload and re-test the request code rather than assuming a like-for-like swap.
Related
See migrating: Haiku 3 to Haiku 4.5 for the first step, migrating: Haiku 4.5 to Haiku 5.5 for the second, and claude-haiku-5-5 for the current Haiku.
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).Could not confirm: Anthropic's current pricing and caching pages no longer list this model, so its prices and cache minimum are the last figures we held; we could not confirm whether any partner platform still serves it.
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.