# Release 2026.18 ### **📌 Release Overview** - **2026.18** - **Unique hosted:** April 29, 2026 - **Self hosted:** May 6, 2026 --- # Chat Platform ### **🚀 What’s New (Business/End-User Focus)** User **New Version Modal**: **Smoother Chat releases:** Chat is deployed with a **rolling update** strategy so new versions roll out gradually instead of replacing every instance at once—releases are less likely to take the whole app offline for a moment. **Stay on the latest build after a release:** When `FEATURE_FLAG_DEPLOYMENT_VERSION_CHECK` is on, if you keep Chat open while a new version goes live, you get a **“New version available”** prompt to **reload** so your session matches the new deployment and avoids odd behavior from mixed old/new versions. Admin **Enable Feature flag**: `FEATURE_FLAG_DEPLOYMENT_VERSION_CHECK` needs to be enabled on `chat` service only. User **File Preview in Select Files to Chat**: When `FEATURE_FLAG_FILE_PREVIEW_UN_19395` is on: Users can now directly preview files through the "Select Files to Chat" sidebar, eliminating the need to search for them in the knowledge base. File types that cannot be rendered in the browser will be downloaded instead. User **Select Uploaded Files as Chat Context:** When `FEATURE_FLAG_ENABLE_SELECTED_UPLOADED_FILES_UN_18215` and `FEATURE_FLAG_ENABLE_NEW_ANSWERS_UI_UN_14411` are on, a new **Uploaded files** tab now appears in the chat's Select Files side panel, listing files uploaded in the current chat. Users can include or exclude individual files (or all at once via select-all) to control which uploaded files are used as context for their next message. ### **🔧 Improvements** User **PDF Viewer Opens at Readable Zoom:** Clicking a source reference in chat now opens the PDF side panel at 100% zoom in a wider default panel, so documents are immediately readable without manual zooming or resizing. Users can still adjust zoom and panel width as before. User **Hallucination Check Indicator on Low-Risk Messages:** A subtle indicator (a small green check icon inline with the message actions) with the assessment title on hover, now appears on assistant responses that passed the hallucination check (low/green). Moderate (yellow) and high (red) results continue to surface as the prominent floating bubble above the chat input, so successful checks are visible without competing with the more attention-grabbing warnings. ### **🐞 Bug Fixes** User **Advanced formatting in chat history:** Fixed "Failed to load" errors when viewing past chat search results containing advanced formatting blocks (charts, HTML rendering), and removed raw fence markup from chat history sidebar previews. User **AdminPurple MCP Route Guard feature flags**: Fixed feature flag evaluation in the MCP Route Guard to read flags from FeatureFlagContext instead of direct config access, ensuring flags are correctly resolved per-tenant. # RAG ### **🔧 Improvements** User **Exclusive** **File Access(Experimental):** Improved folder privacy so if you only have access to certain files within a folder you do not have any access to, you'll now see `restricted` or whatever you have configured as the default instead of other folder names you don't have permission to view. # Agentic Table ### **🔧 Improvements** User **Efficient export to the answer library:** Row metadata is fetched together with cell data when you export a sheet to the answer library, instead of one request per row—large sheets see fewer background calls. Admin **RfpAgent on dedicated service**`assistants-agentic-table` : The RFP agent runs in this service; The Chat backend now targets its webhook instead of routing everything through the `assistants-core` backend service, isolating Agentic Table workloads for safer upgrades and tuning. # Unique AI ### **🐞 Bug Fixes** User **Fix LaTeX Rendering**: Fixed an issue where LaTeX-style outputs (e.g., from Qwen3) were not rendered correctly in the frontend and appeared with raw `$$` delimiters. LaTeX expressions are now properly rendered, consistent with Markdown behavior across supported models. Admin **New Models Added to Unique AI Approved List**: GPT-5.5, Anthropic Claude Opus 4.6, and Anthropic Claude Opus 4.7 have been added to the approved model list. These models are now appearing in the Unique AI dropdown, provided they are enabled for your organization. ### **Deprecation Notice:** Admin **Internal Search Scope IDs:** The `scopeIds` parameter in the internal search configuration is deprecated. This parameter should no longer be used for any new or future spaces. If you are currently using `scopeIds` to limit search scopes, please migrate these settings to the Knowledge Base filter available at the top of the internal search configuration. Existing Spaces will continue to work, but should be migrated to the new approach. **Deprecated parameter** **Use filter to limit knowledge base available for search** # Language Model ### New Model Deployments Admin **GPT-5.5 is now available both through Azure and LiteLLM:** For Azure OpenAI, use `AZURE_GPT_55_2026_0424`, for OpenAI via LiteLLM, use either `litellm:openai-gpt-5-5`. Note that currently the model is only available with Global deployment. Admin **Claude Opus 4.7 is now available both throughLiteLLM:** Use either `litellm:anthropic-claude-opus-4-7`. # Upcoming Model Retirement: ### **Claude Opus/Sonnet 4** Admin Due to the upcoming retirement of **Claude Opus 4 and Claude Sonnet 4**, all spaces and AI templates previously using `litellm:anthropic-claude-4-sonnet` and `litellm:anthropic-claude-4-opus` have been migrated to **Claude Sonnet 4.6** (`litellm:anthropic-claude-sonnet-4-6`) and **Claude Opus 4.6** (`litellm:anthropic-claude-opus-4-6`).