Haiku 4.5 Left Claude’s Model Menu: What Users and Developers Should Do

TL;DR: On October 29, Anthropic removed Haiku 4.5 from the model selector in chat, Cowork and Design for Free, Pro and Max users; existing chats stay, but their next response uses Haiku 5.5.^[1] Claude Code, Team and Enterprise plans, and the Claude Platform API retain Haiku 4.5 access. Haiku 5.5 launched on October 7, so this is a product-menu change, not a new model launch.^[1][2]

If Haiku 4.5 vanished from your Claude menu, the model itself has not vanished. Anthropic announced the change on October 15 and put it into effect on October 29 for chat, Cowork and Design on Free, Pro and Max, including Desktop and Mobile.^[1]

An individual choosing a model in chat now gets Haiku 5.5. Developers can still call Haiku 4.5 through the API; Anthropic lists its API ID as active.^[1][8] Treating a menu removal as an API shutdown would trigger a rushed migration for the wrong reason.

Cover: generated editorial illustration of separate access routes. It represents product choices, not a Claude interface screenshot.

The Scope: A Menu Removal, Not a Model Retirement

Anthropic says Haiku 4.5 no longer appears under “More models” for affected Free, Pro and Max experiences. If it was selected, Claude switches the selection to Haiku 5.5. Team and Enterprise plans, Claude Code and the Claude Platform API are explicitly outside this removal.^[1] A Team or Enterprise administrator can still control which models members see, so organization policy may narrow the options in a particular account.^[5]

Where you work	Haiku 4.5 after October 29	Practical move
Free, Pro or Max chat, Cowork or Design	Removed from the selector; Haiku 5.5 replaces it[1]	Check the model name before sending a sensitive or repeatable task.
Existing Haiku 4.5 chat on those plans	Chat remains; its next answer uses Haiku 5.5[1]	Preserve a copy if the historical record matters; compare new output under the new model.
Claude Code	Haiku 4.5 remains available under this change[1]	Use /model or session configuration to select deliberately.[7]
Team or Enterprise	Haiku 4.5 remains available under this change[1]	Ask an admin if a permitted model is absent from your menu.[5]
Claude Platform API	Haiku 4.5 remains active[8]	Keep the current ID if it still meets requirements; test before changing it.

The real story isn't model extinction. Anthropic is setting a new consumer default while keeping an older-model path for organizations and developers. That changes what a continued chat means.

Existing Chats: The Transcript Stays, the Next Answer Changes

Anthropic says Haiku 4.5 chats have not been deleted. Continue one after October 29 on an affected plan and the next response comes from Haiku 5.5. You can change models at any point in a chat, with the change applying from the next response; the model control sits next to the send button.^[1][5]

For a repeated workflow, record the date, selected model and a representative output before judging whether behavior changed. A preserved transcript is not a frozen execution environment. Individual Free, Pro and Max users can export chat history from Settings > Privacy in the web app or Claude Desktop. Export is unavailable from iOS and Android, and cannot be imported into another personal account.^[6]

Let's be clear: exporting archives chats; it does not keep Haiku 4.5 answering in the affected menu.^[1]

Plan Limits: Do Not Promise a Fixed Message Count

Haiku 5.5 is the replacement in affected menus, but a model switch does not give every user the same number of messages. Anthropic says usage depends on conversation length, complexity, features, model and effort. Paid plans have five-hour session limits and weekly limits; Pro's weekly allowance applies across models. The next reset time appears in Settings > Usage.^[9][10]

Haiku 5.5 has an effort control. Anthropic recommends Low or Medium for routine work to stretch usage; higher effort takes more tokens and reaches limits faster. In Claude's chat interface, thinking cannot be turned off for Haiku 5.5.^[5] For routine tasks, lower effort and shorten context before paying for a higher tier. For harder tasks, test another model available to your plan; its usage cost may differ.

What's often overlooked is the difference between a subscription limit and API billing. Claude Platform calls have token prices, not the consumer chat plan's message allowance. Anthropic lists Haiku 5.5 at $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens; above that prompt threshold the rates are $0.50 and $2.50. Haiku 4.5 is listed at $1 and $5.^[2][4] The lower unit price does not guarantee an identical bill because token counts and request behavior change.

API Migration: Price Is Only One Line in the Diff

Anthropic launched Haiku 5.5 on October 7 for high-volume, latency-sensitive work.^[2] Its Claude API ID is claude-haiku-5-5, with no date suffix or separate alias; Haiku 4.5 uses claude-haiku-4-5 or claude-haiku-4-5-20251001.^[3] The October 29 menu decision does not require an immediate API change.

Run the comparison with the same prompts and acceptance criteria, then check the integration points most likely to fail:

1. Recount tokens and output headroom. Anthropic says the same text is approximately 30 percent more tokens with Haiku 5.5's tokenizer. Revisit budgets, max_tokens and prompts over 100,000 tokens; that threshold changes the per-token price.^[3]
2. Update thinking and response parsing. Replace thinking: {"type":"enabled","budget_tokens":N} with adaptive thinking or leave the setting unset, then choose effort. The old setting returns a 400 error. Read content blocks by type, since thinking can precede text; allow enough max_tokens for both.^[3]
3. Remove incompatible request patterns. Haiku 5.5 rejects temperature, top_p, top_k and final assistant-turn prefill. Its computer-use tool version also changes, with different requirements by platform. Test those requests rather than only changing the model string.^[3]
4. Test refusal and conversation replay. The migration guide says to handle stop_reason: "refusal", preserve earlier turns when replaying thinking blocks, and replay those blocks through the account that produced them. A successful HTTP response can still omit cross-account thinking history.^[3]

Here’s the genius of keeping Haiku 4.5 on the API for now: teams can run a controlled A/B migration instead of inferring production behavior from a consumer menu. The uncomfortable truth is that a cheaper published token price says little about total spend until you count the new tokens, long prompts, retries and accepted outputs. Anthropic's claim that Haiku 5.5 costs about 75 percent less on average is a vendor estimate, not a guaranteed saving for a particular workload.^[2]

The Decision: Match the Surface to the Work

For individual chat users, verify the selected model and keep working with Haiku 5.5 unless a specific task shows a quality or usage problem. For Claude Code users, choose explicitly with /model when reproducibility matters. For API owners, leave Haiku 4.5 in place while a test suite compares output quality, token counts, error handling and total cost against Haiku 5.5. Anthropic's status page still calls the Haiku 4.5 API model active, but its stated earliest retirement date has passed; that is a reason to monitor future notices, not to invent a retirement announcement.^[7][8]

The model selector is now a product policy surface as consequential as a release note. A transcript can survive while the model behind the next answer changes. The teams that treat model identity as a documented input, rather than a background preference, will be the ones that notice the difference before their users do.

Last updated: October 30, 2026