n8n 2.42 and 2.43: what's new and should you upgrade

n8n 2.42 (released September 29, 2026) adds prompt caching to the Anthropic node and renames Azure OpenAI Chat Model to Azure AI Foundry Chat Model. n8n 2.43 (October 6) makes Browser Use in n8n Assistant available to everyone and makes the Migration Report easier to act on before n8n 3.0. For production, run the current stable release, 2.42.6; 2.43 is still on the beta track.

The short version

• Which version to run: on October 11, 2026, npm's stable tag points to 2.42.6 and beta to 2.43.3 (both published October 9).
• 2.42 highlights: Anthropic prompt caching, the Azure AI Foundry rename, SSO first on the sign-in page, Microsoft Teams activity notifications, and Gateway credits for agent web search.
• 2.43 highlights: Browser Use for all users, Migration Report impact categories and a progress bar, an Azure deployment dropdown, Supabase OAuth2, and a redesigned Execute button.
• Action needed: reconnect Microsoft Teams OAuth2 credentials if you want activity notifications, and decide whether Browser Use should stay on for your instance.

All details below come from n8n's release notes.

Should you upgrade?

n8n describes the two release tracks this way: stable "is for production use," while beta "is the most recent release" and "may be unstable." Stable is a later patch of a release that has already spent time on the beta track.

So the practical answer is:

• Production instances: move to 2.42.6 if you are on an older version and want prompt caching or the SSO changes. Pin the exact version in Docker Compose rather than using a moving tag.
• Test or staging instances: try 2.43.3 to check the new Migration Report against your workflows before n8n 3.0.

services:
  n8n:
    image: docker.n8n.io/n8nio/n8n:2.42.6

n8n 2.42: what changed

Prompt caching in the Anthropic node

The Anthropic node's Text > Message operation gets a Prompt Caching option: Disabled, 5 Minutes or 1 Hour. This matches the option the Anthropic Chat Model sub-node already had. With caching on, large system prompts and repeated tool-call loops are billed at Anthropic's reduced rate for cached tokens.

It is off by default, so existing workflows don't change. Reported token usage now also includes cache creation and cache read tokens, which makes cost tracking more accurate.

Azure OpenAI Chat Model is now Azure AI Foundry Chat Model

The node was renamed because it supports the whole Azure AI Foundry catalog, including Llama, Mistral, Cohere, DeepSeek, Grok and Claude models. Existing workflows are not affected, and searching for "Azure OpenAI" still finds it.

SSO comes first on the sign-in page

When SAML or OIDC is your login method, the sign-in page now leads with Continue with SSO, and the email and password form is collapsed. After login, users now return to the page they originally opened, including with OIDC. Instances without SSO see no change.

Microsoft Teams: activity notifications and meeting attendees

• A new Activity Notification resource with a Send operation sends activity feed notifications to a user. Existing OAuth2 credentials must be reconnected to get the new scope.
• Online Meeting Create, Create or Get, and Update now support attendees. Looking up people requires the User.Read.All permission.

Agents and AI

• Agent web search with Gateway credits: when Brave Search is served through n8n's AI Gateway, agents can pay for searches with Gateway credits instead of your own Brave API key. SearXNG is not supported this way.
• Agent Builder: a write lock so only one browser tab edits an agent at a time, queued messages in the preview chat, a revamped skills modal with file upload, and a Test agent label in place of the old preview action.
• Mistral Vibe joins the Connect a client picker under Settings > Instance-level MCP.

New to agents? Our explainer covers how n8n Agents differ from the AI Agent node.

Admin and governance

• Instance owners and admins now appear in every project's member list with "full access." Project member lists also gain search and pagination.
• With granular credential sharing on (N8N_ENV_FEAT_CRED_SHARING), the credential picker groups credentials by where they come from, such as Available to you or Available to everyone.
• When a node or credential type policy blocks a save, publish or manual run, the editor now lists each blocked type and its scope. This requires the type availability policies module and a license that includes it.

n8n 2.43: what changed

Browser Use is now available to all users

Browser Use in n8n Assistant is no longer an experiment. Instance admins control it with N8N_INSTANCE_AI_BROWSER_USE_ENABLED, which is on by default. It isn't available on phones or tablets, including iPads in desktop mode.

If you don't want it on your self-hosted instance, set the variable explicitly:

services:
  n8n:
    environment:
      - N8N_INSTANCE_AI_BROWSER_USE_ENABLED=false

The same release turns off the experimental "Set up automatically" credential option by default.

A Migration Report you can prioritize

This is the most useful change if you are preparing for n8n 3.0. The Migration Report in Settings replaces low, medium and critical severities with four impact categories:

• Upgrade blocked
• Executions fail
• Behavior changes
• Capability removed

Rules are sorted by impact, with blockers first. A progress bar at the top shows how many of your workflows are compatible with the target version ("X of Y compatible"). Start with Upgrade blocked and Executions fail, then work down. Our n8n 3.0 breaking changes guide explains what each change means and how to fix it.

Node updates

• Azure OpenAI Chat Model: pick a deployment from a dropdown instead of typing its name, which avoids DeploymentNotFound errors. New nodes use version 1.1. The list needs an Azure AI Foundry credential with a project set; version 1 nodes keep their text field.
• Supabase: OAuth2 credentials alongside API keys. The same credential can enable the Supabase MCP server in n8n Assistant.
• Databricks: the Job resource is now visible, with Run, Get Job, Get Run and Get Run Output.
• Crypto: asymmetric encrypt and decrypt can use SHA-1 for RSA-OAEP. SHA-256 stays the default.

Editor changes

• The canvas button now just reads Execute and shows the selected trigger's icon. Picking a trigger from its menu runs the workflow right away.
• n8n Assistant gets closable tabs that are saved per thread, a button to open any workflow, data table or agent in the project as a tab, and Cmd+J / Ctrl+J to open and close the panel.
• NODES_EXCLUDE can now hide a generated AI tool variant (for example microsoftOutlookTool) without removing its base node.

A note on n8n 3.0

n8n's 3.0 breaking changes page now opens with "n8n 3.0 has been released." As of October 11, 2026, though, npm lists no 3.x version, and the newest releases are 2.42.6 and 2.43.3. Treat 3.0 as imminent and use the time to clear your Migration Report. We will cover the release when it ships. For all release coverage, see the releases tag.