Claude Sonnet 4.5 retires on 30 November 2026

Updated 2026-10-05 · 5 min read · By the Convo360 team

What Anthropic has actually announced

On 30 September 2026 Anthropic announced the deprecation of Claude Sonnet 4.5, the model published as claude-sonnet-4-5-20250929, with retirement from the Claude API scheduled for 30 November 2026. Anthropic recommends migrating to a current model before that date.

It is worth being precise about the two words, because they mean different things and only one of them is urgent. Deprecated means the model still works but is no longer recommended, and no further improvements will be made to it. Retired means the model id stops being served. After 30 November, a request naming that id does not fall back to something newer. It returns an error.

That distinction matters because deprecation notices are easy to ignore. Nothing visibly changes on the day a model is deprecated, so the announcement reads like news rather than a task. The consequence lands weeks later, on the retirement date, usually on a working day when nobody is expecting it.

Why a retired model takes a bot down completely

People reasonably assume that an old model will keep working a little worse, or that the provider will quietly substitute its replacement. Neither happens. The model id is part of the request, and once the provider stops recognising it the request is rejected outright.

For a chatbot or a voice agent, that failure is total rather than partial. Every incoming message needs a model reply, so if the model call fails, every conversation fails. A customer messages your WhatsApp number and gets silence. From the outside this looks exactly like the product being broken, and it is the sort of outage that gets discovered by a customer rather than by you.

This has already happened in the Indian market with voice. When Sarvam retired its bulbul:v2 text-to-speech model, every call on that voice failed at once, and most of the businesses affected found out when a caller complained. The announcement had been published. Nobody had read it.

Check whether your bot is actually affected

Before changing anything, find out whether this applies to you at all. Plenty of bots are pointed at a default model and were never set to Sonnet 4.5 in the first place.

• Open each of your agents and look at the model field. You are looking for any id beginning claude-sonnet-4-5.
• Check every agent, not just the main one. Test agents and old drafts use real keys and will fail just as loudly.
• Check anything calling the Claude API outside your bot, such as a script, a spreadsheet add-on or an internal tool.
• If you pin a dated model id anywhere in code, search for the full string claude-sonnet-4-5-20250929.

In Convo360 the model is a per-agent setting, so this is a visual check on the agent screen rather than a search through code. Each agent names the provider and the model it uses, which also means a workspace running several agents can be on several different models at once.

Choosing what to move to

Anthropic released Claude Sonnet 5.5 on 28 September 2026, and for most support and sales bots it is the natural replacement. It sits in the same place in the range as Sonnet 4.5 did, which is what you want in a migration: the nearest thing, not the newest thing.

If your workload is genuinely heavier, Claude Opus 5.5 was released on 22 September 2026 with a one million token context window by default and a 128,000 token maximum output. That capacity is useful for long documents and multi step agentic work. It is more than a WhatsApp reply needs, and it costs more per token, so it is usually the wrong trade for a chat bot.

There is also no rule that says you must stay with the same provider. If you were already considering OpenAI or Google Gemini, a forced migration is a reasonable moment to compare them. Our guide to choosing between OpenAI, Claude and Gemini sets out the trade offs.

How to switch without breaking anything

The switch itself takes a minute. The testing is what stops it becoming a second incident.

On a bring-your-own-keys platform like Convo360 the model is a setting on the agent, not something buried in code, so you change it in a dropdown and save. Your Anthropic key does not change, your flows do not change, and your knowledge base does not change.

What can change is behaviour. A newer model often writes slightly longer, formats differently, or interprets a brief instruction more literally. For a bot whose replies go out on WhatsApp, where length is visible and tone matters, that is worth a look before thousands of customers see it.

So after switching, run one real conversation end to end. Ask the three questions your customers actually ask most. If the replies are longer than you want, tighten the system prompt rather than reverting the model, because reverting is only available until 30 November.

Making sure the next one does not surprise you

This will happen again. Between June and October 2026 alone, OpenAI deprecated several model families and a set of text to speech models, Google restricted access to its Gemini 2.5 models, and Anthropic retired Sonnet 4.5. On a bring-your-own-keys platform these decisions are made by someone else and announced on a changelog page.

The practical answer is not to read four changelogs every week. It is to be told when something is removed or repriced, and to ignore the rest. Convo360 reads the official Meta, OpenAI, Anthropic and Google changelogs several times a day and emails the platform owner immediately when an entry removes a feature or changes pricing, while everything merely interesting waits for a weekly digest.

If you would rather set that up yourself, subscribe to each provider's deprecations page specifically, not its blog. Company news and model retirements are published in different places, and the retirements are the ones that take a bot offline.

Step by step

1. Find every agent using Sonnet 4.5Open each agent and look for a model id starting claude-sonnet-4-5, including test agents and old drafts.
2. Pick the replacement modelChoose Claude Sonnet 5.5 for most support and sales bots. Consider Opus 5.5 only for long document or agentic work.
3. Change the model on the agentSwitch the model in the agent settings and save. Your API key, flows and knowledge base stay as they are.
4. Run one real conversationAsk the three questions your customers ask most and check reply length and tone before customers see them.
5. Tighten the prompt if neededIf replies are longer or less direct than before, adjust the system prompt rather than reverting the model.
6. Set up change alertsSubscribe to each provider's deprecations page, or use a platform that emails you when a model is retired or repriced.

Example conversation

Frequently asked questions

Related reading

Available across India — see WhatsApp & AI calling automation by state and city.