Claude in October: Sonnet 4.5 retires on 30 November, Haiku 5.5 is new

27 October 2026 Michael Engel General Costs

Two new models, a halved cache price and a retirement with a deadline: what to switch in your workflows before 30 November.

Source: Model deprecations (Claude Platform Docs, Anthropic), 30 September 2026

Note: I use Claude and Claude Code for my own work. Anthropic does not pay me, and the links are not affiliate links.

Technical drawing: a timeline from September to December 2026 with three model lanes. The Sonnet 4.5 lane turns dashed and ends with a cross at the 2026-11-30 line; before that, an arrow leads to the Sonnet 5.5 lane, which carries a "Cache -50%" marker, and the Haiku 5.5 lane starts below.

What happened: According to Anthropic’s release notes, several things changed for Claude:

28 Sept: Claude Sonnet 5.5 is released.
30 Sept: Claude Sonnet 4.5 is deprecated. It will be retired on 30 November 2026, after which requests fail. The recommended replacement is Sonnet 5.5.
7 Oct: Claude Haiku 5.5 is released, the small model for high-volume work. On Sonnet 5.5, the price of cache reads is halved, meaning stored parts of a prompt that are reused.
21 Oct: The new API version 2026-10-21 rejects temperature, top_p and top_k with a 400 error, even at their default values. New major SDK versions use it automatically.

List prices in US dollars per million tokens (the chunks of text that usage is billed by):

Sonnet 5.5: input 2, output 10, cache reads 0.10 (previously 0.20)
Haiku 5.5, prompts up to 100,000 tokens: input 0.10, output 0.50, cache reads 0.01
Haiku 5.5, longer prompts: 0.50 / 2.50 / 0.05

What this means: A new model name alone is not always enough. According to the migration guide, Sonnet 5.5 rejects, among other things, fixed thinking budgets (budget_tokens), the disabled thinking mode, temperature, top_p, top_k, prefilled assistant responses (prefill) and forced tool use with a 400 error. Depending on the content, the same text also produces about 30% more tokens than on Sonnet 4.5.

What you can do:

Find your usage: In the Claude Console, go to Usage and click Export; the CSV shows usage by API key and model. Then search n8n nodes and scripts for claude-sonnet-4-5.
Switch and test before 30 November: Set claude-sonnet-5-5, remove or replace the settings listed above, test with real tasks and compare the cost per task.
Consider Haiku 5.5 for high-volume work: According to Anthropic, it suits repetitive tasks such as summaries and classification. When moving from Haiku 4.5, also remove or replace budget_tokens, temperature, top_p, top_k and prefill.
Update SDKs only after this clean-up: Older major SDK versions stay on the previous API version.

Context: The deadline applies to Anthropic-operated platforms; Amazon Bedrock and the Google-managed Claude offering set their own dates. “30%+ faster” (Sonnet 5.5 compared with Sonnet 5) and “around 75% less to run” (Haiku 5.5 compared with Haiku 4.5) are vendor claims.

Bringing AI into your own team? That’s what my workshops are for. See the workshops