2026.3.1

Release date: April 15, 2026

Model updates

The following model changes take effect in this release:

• Main and image model: Upgraded from Claude Sonnet 3.7 to Claude Sonnet 4.5
• Fast model: Upgraded from GOT-4o mini to GPT-4.1 mini
• Output token limit: Increased from 8,000 to 16,000 tokens to support longer responses

Starting Prompt button

A Starting Prompt button is now available in the chat input bar for active conversations. Previously, starting prompts were only shown on an empty conversation. You can now access them at any point in a conversation to quickly reuse or share common prompts.

HTML preview

HTML code blocks in Autopilot for Everyone now display a rendered preview alongside the raw markup. This brings parity with the HTML preview behavior already available in Conversational Agents.

Automatic retries

Autopilot for Everyone now retries requests automatically when the LLM gateway returns a transient failure. This reduces the impact of intermittent connectivity issues on response delivery.

Bug fixes

• Fixed an issue where the Content Grounding index was not queried for specialized autopilot configurations.
• Fixed an issue where Autopilot returned HTML code using blockquote formatting instead of a code block, causing it to display incorrectly.
• Fixed an authentication reconnection loop where repeated failures prevented re-establishing a connection. Exponential backoff is now applied between reconnection attempts.
• Fixed an issue where the Clipboard AI panel refreshed too aggressively on focus gain. The panel now refreshes on mouse-over of the Clipboard AI button.
• Fixed an issue where the Personal Workspace dropdown was disabled when an automation card had multiple connectors, each with a single connection.
• Fixed an issue where links to images returned by automations were displayed as plain text instead of clickable hyperlinks.