Can you still use the Claude Haiku 4.5 API?
Yes. As of November 8, 2026, Anthropic’s model page says Haiku 4.5 is still available, but labels it “Legacy” and recommends considering migration to Haiku 5.5. The page lists retirement as “Not sooner than October 15, 2026”—a lower bound, not a promise of support until a particular date. For a new integration, weigh that legacy status against the current model; for an existing integration, check your actual API access and plan a migration rather than treating Haiku 4.5 as a long-term default. (platform.claude.com)
Anthropic says the model’s lifecycle dates apply to Anthropic-operated platforms; Amazon Bedrock and Google Cloud’s Google-managed models set their own schedules. Confirm availability with the provider you plan to use. (platform.claude.com)
Direct Anthropic API: model ID and prerequisites
For Anthropic’s direct API, the documented pinned model ID is claude-haiku-4-5-20251001. The alias claude-haiku-4-5 resolves to that snapshot. Use the pinned ID when you want your configuration to name the specific version, and confirm it is callable from your account. (platform.claude.com)
You need a Claude Console account and an API key, or configured Workload Identity Federation. The direct Messages API endpoint is POST /v1/messages on api.anthropic.com. The cURL example below uses the anthropic-version: 2026-10-21 header shown in Anthropic’s current API documentation. (platform.claude.com)
Treat the API key as a secret. Keep it out of browser code and public repositories; supply it to the command from an environment variable. For example: export ANTHROPIC_API_KEY='your-key'. (platform.claude.com)
Make your first request with cURL
curl https://api.anthropic.com/v1/messages \
-H "x-api-key: $ANTHROPIC_API_KEY" \
-H "anthropic-version: 2026-10-21" \
-H "content-type: application/json" \
-d '{
"model": "claude-haiku-4-5-20251001",
"max_tokens": 1024,
"messages": [
{"role": "user", "content": "Reply with exactly: API connected"}
]
}'
This example follows Anthropic’s documented cURL header and body pattern. The Messages API accepts a user message with string content; its request fields include max_tokens and messages. This example is documentation-based, not a report of a live test. (platform.claude.com)
A response includes a content array of blocks. Read the text value from a block whose type is text rather than assuming a particular block position. The Messages API is stateless: to continue a conversation, send the relevant prior user and assistant messages again with the next request. (platform.claude.com)
Limits, features, and pricing
Anthropic’s Haiku 4.5 overview lists a 200,000-token context window and 64,000-token maximum output. It describes text and images as inputs and text as output. Follow the API’s documented image-content format if you send images. (platform.claude.com)
The model overview lists these direct API prices in USD per million tokens (MTok): (platform.claude.com)
| Usage | Listed price |
|---|---|
| Input tokens | $1 / MTok |
| Output tokens | $5 / MTok |
| 5-minute prompt-cache write | $1.25 / MTok |
| 1-hour prompt-cache write | $2 / MTok |
| Cache read | $0.10 / MTok |
| Batch API | 50% discount on input and output |
Check Anthropic’s official pricing documentation before budgeting or launching because prices can change. The listed Batch API discount is for input and output; the model overview does not say it also discounts cache charges. (platform.claude.com)
Illustrative cost: a request using 1,000 input tokens and 200 output tokens would cost about $0.002 at those listed base rates: 1,000 × $1 / 1,000,000, plus 200 × $5 / 1,000,000. This is a hypothetical calculation, not a measured request, and excludes caching, batch processing, and any other applicable charges.
Haiku 4.5 supports manual extended thinking, not adaptive thinking; Anthropic lists default effort as unsupported. Its model page identifies thinking.type: "enabled" and mentions budget_tokens, but the basic request above leaves thinking off. Consult the current thinking reference before adding those fields. (platform.claude.com)
Should you migrate to Haiku 5.5?
Anthropic’s migration guide covers Messages API code moving from Haiku 4.5 to Haiku 5.5. For Haiku 4.5, its checklist says to replace the model ID, recount prompts, and revisit max_tokens limits and cost estimates. If your requests use manual thinking with budget_tokens, the guide says to change the thinking configuration to adaptive. It also instructs developers to select response content blocks by type, remove temperature, top_p, and top_k, and update assistant-prefill and computer-use patterns where applicable. Treat this as a code migration and test it against your workload; do not assume the new model is a drop-in replacement. (platform.claude.com)
Keep provider-specific instructions separate
The model ID and endpoint above are for Anthropic’s direct Claude API. Amazon Bedrock, Google Cloud, Microsoft Foundry, and API gateways can use different model IDs, credentials, endpoints, billing, and lifecycle terms. For example, CometAPI’s published guide describes its own host and Bearer-token key; those are not Anthropic endpoint or authentication instructions. Use the documentation for the provider you have actually chosen. (platform.claude.com)
Sources
Claude Haiku 4.5 — Claude Platform Docs
Model deprecations — Claude Platform Docs
API overview — Claude Platform Docs
Create a Message — Claude API Reference
Using the Messages API — Claude Platform Docs
API migration guide for version 2026-10-21 — Claude Platform Docs