Anthropic deletes Claude temperature dial in first major API update since 2023

Anthropic just stripped away core sampling controls like temperature, betting that developers should build on models instead of tuning them.

Tools ยท Source: PANews

What happened

Anthropic just dropped Claude API version 2026-10-21. This is the first official API version release in three years. The update completely removes the temperature, top p, and top k parameters from the Messages API. Developers can no longer control model creativity through these dials.

The update brings strict new formatting rules. Passing documents in an assistant turn or citations in a user turn now triggers a 400 error. Anthropic also replaced several deprecated fields in the Admin and Enterprise Analytics APIs. Alongside the API changes, Anthropic released major updates for all its client SDKs, including Python, TypeScript, Go, Java, PHP, Ruby, and C sharp. These tools now default to the new API version. They also switched licensing from MIT to Apache 2.0.

Anthropic is also expanding enterprise access and safety features. They launched Claude API access through Google Cloud, complete with IAM controls, Marketplace billing, and a Fast mode for Claude Opus 5.5. They also introduced user profiles to track end customers. This includes new enforcement status fields to see if Anthropic has restricted or banned a specific user.

Key facts

Why it matters

You lose the ability to tweak how creative or deterministic Claude gets. Anthropic is taking away the dials. They want you to focus on prompting and product architecture instead of fiddling with sampling parameters. The old 2023-06-01 API version still works for now. However, Anthropic explicitly stated they will ship breaking changes more frequently moving forward. Developers must adapt to incompatible changes before new models require them.

The second order effect is a massive shift in enterprise risk management. The new user profile endpoints let platforms isolate bad actors. If an end user violates safety rules, Anthropic can ban that specific user profile instead of shutting down your entire organization API access. Furthermore, the shift to Apache 2.0 for all SDKs gives commercial builders more legal clarity. Enterprise admins also gain granular control over web fetch and search domains through the Claude Console, directly impacting how Managed Agents operate.

For builders

Update your API request formatting immediately

The new version throws a 400 error if you put documents in an assistant turn or citations in a user turn. Developers migrating to the new SDKs must rewrite their message payloads. Failure to do so will result in broken integrations.

Isolate user bans with new profiles

Platforms and resellers can now pass user profiles in their API requests. If a customer breaks the rules, Anthropic bans that specific user instead of nuking your entire enterprise account. This protects your core business from rogue users.

Manage web fetch domains for agents

Organization admins can now set allowed or blocked domains for web fetch and search. Managed Agents will strictly follow these domain restrictions. You must configure these lists to prevent agents from accessing unauthorized external data.

My take

Taking away the temperature dial is a massive flex. Anthropic is telling developers their base models are good enough out of the box. Stop tweaking parameters and start building actual products. They took your dials but gave you open source SDKs and granular enterprise controls.

Original reporting: PANews. This is my rewrite and opinion.

More AI news for builders