IT Center Changes

Bug Fixes and Improvements:

• Improved processing of tool calls in the Responses endpoint.
• Admins of organizations without a user group can now view the front-end application again.

Bug Fixes and Improvements:

• Improved the design of modal dialogs, especially on mobile devices.
• Setup administrators who are not members of any active group can now still select all groups.
• Fixed an issue in the API key modal where the group name could be displayed in the wrong language.
• Fixed an issue that prevented Hard Limit notification emails from being sent correctly.

Features:

• Added support for the new model: gpt-image-2

Bugfixes and Improvements:

• Improved the overall modal design, especially for mobile devices
• Enhanced the design of the error modal
• Image generation settings are now persisted and remain available after a page reload
• The deployment list in the admin area now directly indicates whether a deployment is enabled for the frontend and/or API
• Free models can now continue to be used even after the hard limit has been reached
• Fixed an issue where chat titles would sometimes only appear after reloading the page

Features:

• Support for new models added: gpt-5.4 and gpt-5.5
• Reports can now be exported as JSON or CSV
• Tiered pricing based on input tokens is now supported
• Input tokens can now be limited separately per deployment for frontend and API

Bugfixes and Improvements:

• Image models now return an error message if generation fails due to the content filter
• The response endpoint now returns an error message when the context window is exceeded

Bugfixes and Improvements:

• Some models that were temporarily unavailable for admins are now selectable again: glm-4.7, apertus-70b-instruct-2509, and deepseek-r1-distill-llama-70b.
• An issue with permission checks for creating shared chats has been fixed.

Bugfixes and Improvements:

• Opening a chat no longer triggers a full page reload
• For OpenAI models hosted via GWDG, only the supported chat completions endpoint is now available for selection
• Text overflow issues with long words and URLs have been fixed
• Generated images are now displayed directly and no longer require a page reload
• The allowed values for reasoning_effort have been corrected for certain OpenAI models
• Chat links can now be opened directly even when the user is not logged in
• Shared chats can now be opened directly even when the user is not logged in

Features:

• Added support for the /api/v1/responses endpoint
• Added support for the Responses endpoint in the chat frontend
• Added support for new models:
  • gpt-5.2-codex, gpt-5.3-codex, gpt-5.4-mini, gpt-5.4-nano, and the latest GWDG models

Bugfixes and Improvements:

• Improved mobile sidebar and user menu
• Improved API validation and error handling, particularly for incorrect model types and provider errors
• Improved token calculation performance
• Fixed an issue with custom Tailwind color names after a package update

Bugfixes and Improvements:

• The model selection list is now scrollable when it becomes too long
• The model overview has been redesigned for improved clarity

Features:

• Added support for new models:
  • gpt-5.3-chat, Mistral-Small-4-119B-2603

Bugfixes and Improvements:

• No spaces or special characters are allowed anymore in internal deployment names
• Improved performance when accessing the API
• For image generation, the input image is now only displayed if it was manually selected
• Minor improvements to shared chats, especially regarding model selection and chats with images
• Improved behavior when manually cancelling a completion
• Fixed an issue with title generation when it takes longer than the actual prompt
• Fixed a problem with database indexes that prevented the app from starting

Features:

• Added support for new models:
  • Mistral-Large-3-675B-Instruct-2512, gemini-3-flash-preview, gemini-2.5-pro, gemini-2.5-flash, gemini-2.5-flash-lite, gemma-3-27b-it, gemma-3n-e4b-it

Bugfixes and Improvements:

• Title generation is no longer counted toward the maximum number of parallel requests per user
• Added support for the encoding_format parameter in the embedding endpoint
• Improved parameter selection behavior for image generation
• Restricted access to own chats in user groups that are no longer accessible