When is Claude Sonnet 4.5 being retired?

Anthropic retires claude-sonnet-4-5-20250929 on November 30, 2026. It was deprecated on September 30, 2026. After the retirement date, requests to the model fail. The date applies to Anthropic-operated platforms: the Claude API, Claude Platform on AWS, Claude Models on Google Cloud (Anthropic managed) and Microsoft Foundry; Amazon Bedrock and Google-managed Claude on Google Cloud set their own schedules.

What should I use instead of Claude Sonnet 4.5?

Anthropic's recommended replacement is Claude Sonnet 5.5 (claude-sonnet-5-5). It is also cheaper: $2 input and $10 output per million tokens with $0.10 cache reads, against $3 and $15 with $0.30 cache reads for Sonnet 4.5. If you need the smallest code change, Claude Sonnet 4.6 is still active (not retired before February 17, 2027) at the same $3/$15 price, but it only buys time. For cheap high-volume work, Claude Haiku 5.5 costs $0.10/$0.50 for prompts up to 100,000 tokens.

What breaks when moving from Sonnet 4.5 to Sonnet 5.5?

Five settings that Sonnet 4.5 accepted return a 400 error on Sonnet 5.5: assistant message prefill, thinking budgets (budget_tokens), sampling parameters (temperature, top_p, top_k), forced tool choice, and thinking set to disabled. Sonnet 5.5 also thinks by default, takes an effort parameter, and uses a newer tokenizer that produces about 30% more tokens for the same text, so recheck max_tokens and cost estimates.

How do I find out if I still use Claude Sonnet 4.5?

In Claude Console, open the Usage page, click Export and review the CSV, which breaks usage down by API key and model. Search your code and configuration for claude-sonnet-4-5 as well, including model aliases in SDK wrappers, gateways and agent frameworks.

Quick Answer

Claude Sonnet 4.5 Retires Nov 30, 2026: What to Use Instead

Published: November 8, 2026

The short answer

Claude Sonnet 4.5 (claude-sonnet-4-5-20250929) stops working on the Claude API on November 30, 2026. Anthropic deprecated it on September 30, 2026 and names Claude Sonnet 5.5 (claude-sonnet-5-5) as the replacement. Sonnet 5.5 is cheaper — $2/$10 per million tokens versus $3/$15 — but five request settings that Sonnet 4.5 accepted now return errors, so test before you switch.

Your options

Model	Status (Nov 8, 2026)	Price per 1M tokens (input / output / cache read)	Migration effort	Use it for
Claude Sonnet 5.5 (recommended)	Active; not retired before Sep 28, 2027	$2 / $10 / $0.10	Medium: remove prefill, sampling and thinking-budget settings	Default replacement
Claude Sonnet 4.6	Active; not retired before Feb 17, 2027	$3 / $15 / $0.30	Low: same tokenizer; remove prefill, set thinking adaptively	Stopgap only
Claude Haiku 5.5	Active	$0.10 / $0.50 / $0.01 (prompts ≤100K tokens)	Medium	Classification, extraction, high-volume calls
Claude Opus 5.5	Active	$4 / $20 / $0.20	Medium	Hardest coding and agent work
Claude Sonnet 4.5	Deprecated; retires Nov 30, 2026	$3 / $15 / $0.30	—	Nothing new

Prices from Anthropic’s pricing page on November 8, 2026; see also our current API prices.

Which platforms the date applies to

The November 30 date covers Anthropic-operated platforms: 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) set their own retirement schedules, so check those consoles separately if you call Sonnet 4.5 through them.

Step 1: Find every place you call Sonnet 4.5

In Claude Console go to Usage → Export and read the CSV: it breaks usage down by API key and model. Then search code, environment variables, gateway configs and agent frameworks for claude-sonnet-4-5. Anthropic’s migration guide also suggests letting Claude Code do the edit with /claude-api migrate this project to claude-sonnet-5-5.

Step 2: Remove the settings that now return errors

Anthropic’s Sonnet 5.5 migration guide lists five settings that produce a 400 error on Sonnet 5.5:

Assistant prefill — a conversation must end with a user message. Replace prefill used for output format with structured outputs or instructions.
Thinking budgets — thinking: {"type": "enabled", "budget_tokens": N} is rejected. Sonnet 5.5 thinks by default; control depth with the effort parameter.
Sampling parameters — remove temperature, top_p and top_k.
Forced tool choice — use auto with strict tool definitions instead.
thinking: {"type": "disabled"} — use the documented options for turning off up-front thinking instead.

Sonnet 4.5 had no effort parameter, so set one explicitly. The API default for Sonnet 5.5 is high; medium is cheaper and is what Claude Code and the Claude apps use.

Step 3: Recheck token budgets and costs

Sonnet 5.5 uses a newer tokenizer that produces about 30% more tokens for the same text than Sonnet 4.5. Raise max_tokens where outputs were near the limit and redo cost estimates: the lower per-token price usually still wins, but not always by a third. The minimum cacheable prompt falls to 512 tokens (from 1,024), so more prompts benefit from caching at $0.10 per million. Images can now use a higher-resolution tier up to 2,576 pixels on the long edge, which also costs more tokens.

Step 4: Watch for new safeguards

Anthropic notes that real-time cyber safeguards are new for teams coming from Sonnet 4.5. If you do legitimate security work and see refusals, Anthropic points to its Cyber Verification Program.

Step 5: Evaluate, then cut over before November 30

Run your own eval set against Sonnet 5.5 at medium and high effort, compare quality, latency and cost per task, then switch. Do not leave it until the final days of November: requests to a retired model simply fail. Our guide to running an AI model upgrade safely has a rollout checklist.

Last verified: November 8, 2026.