Claude 3 Haiku became available on March 13, 2024, nine days after the broader Claude 3 announcement. It was positioned for low-latency, high-volume workloads with text and image understanding. Its dated ID `claude-3-haiku-20240307` was retired on April 20, 2026 in Anthropic-operated APIs.
Historical research cutoff: 25 October 2026. Primary-source historical claims are date-scoped; cloud entitlements, region-specific availability, current pricing and rate limits are not implied.
Release identity at a glance
- Provider
- Anthropic
- Model family
- Claude 3 Haiku
- Public launch
- 2024-03-13
- API ID / checkpoint
- claude-3-haiku-20240307
- API retirement status
- 2026-04-20
The release and why it mattered
Launch materials highlighted fast document processing and applications such as customer support. Vendor throughput metrics are not independently measured performance guarantees and depend on prompt size and serving conditions. Haiku’s role was not to match Opus on every reasoning task, but to serve workloads where speed and scale can matter more than maximum model capability.
Capabilities and genuine use cases
Extracting a form field, routing a message or drafting a brief answer are plausible trials. Still, a speed-oriented model can misread a blurred receipt or confuse visually similar numbers. For recordkeeping systems, confidence gates and cross-checks should be applied before writing extracted values to databases.
Integration and version identity
Migrate an old Haiku integration by inventorying the exact model name, prompt templates, input image processing and output validators. Use the vendor retirement schedule for Anthropic endpoints, and separately confirm Bedrock or Google service lifecycle if applicable. A live service with a compatible name may run a different model generation.
Limits and evidence gaps
The 2024 announcement’s price and performance framing should not be converted into 2026 costs. Strong benchmark numbers on selected image categories are not proof of reliable OCR on unseen field documents. The model’s retired status is decisive regardless of how competitive it once appeared.
How to choose or migrate
Haiku is an instructive example of the volume-oriented model tier. The replacement decision should focus on measured latency percentiles, false positives, data-handling constraints and escalation costs—not marketing descriptions of fast inference.
Reproducible evaluation checklist
For older Claude services, compare the vendor-operated API lifecycle with the lifecycle published by the actual cloud partner. Exercise long inputs, structured outputs and tool argument validation as separate acceptance tests. Do not assume that matching family names imply identical backend checkpoints.
Collect the exact Messages API version, system instructions and model string from a reproducible request. An active cloud model and a retired direct-API model can coexist under similar product naming.
Sources & Further Reading
- www.anthropic.com — documentation/reference (supplier-provided reference; not independently checked in this pass)
- www.anthropic.com — documentation/reference (supplier-provided reference; not independently checked in this pass)
- platform.claude.com — documentation/reference (supplier-provided reference; not independently checked in this pass)
Source documentation establishes vendor-published facts, not independently measured model quality, tenant access, contractual rights or current prices. Use a dated pricing ledger and an account-specific entitlement check for deployment decisions.