v7.58

Release Notes

This update introduces new safety features, along with important bug fixes and improvements.

Assistants

New Features

• Introducing support for input and output moderation in Next-Gen Assistants. PingPong will request that OpenAI returns moderation scores for the model input and generated output.
• PingPong does not currently take any action based on moderation results. Moderation scores are stored in full for future use.
• Introducing support for the input_file message type in Next-Gen Assistants.
  • For PDF documents, on models with vision capabilities, such as gpt-4o and later models, the API extracts both text and page images and sends both to the model.
  • For non-PDF document and text files (for example, .docx, .pptx, .txt, and code files), the API extracts text only.
  • For spreadsheet files (for example, .xlsx, .csv, .tsv), the API parses up to the first 1,000 rows per sheet and adds model-generated summary and header metadata so the model can work from a smaller, structured view of the data.

Notes

• The moderation model is free to use and no additional charge is added to the Responses API request when the moderation request is included.
• The moderation score feature in Next-Gen Assistants builds upon existing safety features in PingPong, including the per-user, opaque safety identifier provided with every user message in Next-Gen Assistants in Chat mode, and all Voice mode conversations.
• The input_file message type is currently unavailable for user-uploaded files in Next-Gen Assistants. Continue utilizing File Search and Code Interpreter for PDFs and other documents.

Lecture Slides (Preview)

Note

This release adds new capabilities for Lecture Slides, which remain under active development and are not yet officially supported in Groups. APIs, features, and UI/UX may change before final release.

Lecture Slides mode is currently visible only to institutional admins when creating an assistant.

New Features

• The Lecture Slides editor now includes a slide timeline with PDF thumbnails, slide selection, and question markers so instructors can move through a deck and its Knowledge Checks from one timeline-style view.
• Instructors can now add, edit, and delete Lecture Slide Knowledge Checks directly in the slide editor.
• Lecture Slide Knowledge Checks now support three authoring modes: full instructor-authored questions, partial drafts for the model to complete, and marker-only questions where the instructor chooses the slide break and lets the model write the full question.
• Instructors can now edit Knowledge Check prompts, spoken intros, answer choices, correct-answer selection, and per-answer spoken feedback from the assistant editor.
• Instructor-authored Knowledge Checks can now be included when creating or updating Lecture Slide assistants, instead of relying only on generated questions.
• Lecture Slide generation now preserves complete instructor-authored questions exactly and avoids generating duplicate questions after the same slide.
• Partial Knowledge Check drafts and marker-only question requests are now passed into manifest generation so the model can complete or create questions at the requested slide positions.
• Saving Knowledge Check narration or feedback edits now queues only the affected question audio for regeneration.
• Changing the Lecture Slides voice or requesting audio regeneration now resets both slide narration audio and Knowledge Check audio.
• The assistant save flow now treats Lecture Slide question edits as first-class changes and includes them in the unsaved-changes summary.

Updates & Improvements

• Standardize knowledge check placement between slides for upcoming features.
• Update generation prompts for new knowledge check placement.
• Lecture Slides chat now uses slide narrations, knowledge checks, playback state, and the original slide PDF as context instead of separately generated summary checkpoints and moment contexts.
• The slide processing pipeline now generates only knowledge checks for lecture slide manifests, reducing generation work and keeping chat context closer to the source slides.
• Existing lecture slide PDFs can be backfilled into OpenAI file storage so previously created slide decks can use the same PDF-grounded chat context.
• Lecture Slides chat now provides the model Knowledge Check context in developer messages only as learners reach each check, instead of loading the full list when chat first starts.
• Lecture Video and Lecture Slides now use a shared lecture prompt template with separate context sections for each lesson type.
• New Lecture Slide assistants now start with a slide-specific chat prompt that grounds learner answers in slide narrations, the source PDF deck, current slide state, and knowledge-check history.
• The Edit Assistant page now compares and resets Lecture Slide chat instructions against the Lecture Slides default prompt instead of the Lecture Video default prompt.
• To more effectively manage costs, return to the default service tier, down from priority.
• To more effectively manage latency, use medium reasoning effort instead of high.
• Lecture Slide narration and manifest generation now reuse the stored OpenAI copy of the uploaded slide deck instead of uploading a temporary PDF for each generation step.

Deployment Information

Schema Upgrade	Migration Script	Permissions Update	Task Definition Update	Configuration Update
YES	YES	No	No	No

Deployment Details

• N/A

Related PRs

• feat(slide): move questions to the end of the slide by @ekassos in #1794
• feat(next-gen): intorducing moderation by @ekassos in #1803
• feat(slides): do not generate summaries and moments by @ekassos in #1802
• change(slides): do not include full knowledge check history by @ekassos in #1805
• feat(slides): update default lecture prompt by @ekassos in #1808
• change(slides): lower service_tier and reasoning by @ekassos in #1812
• feat(slides): revamp slide editor and question generation by @ekassos in #1818
• change(slides): use stored slide deck on openai for generation by @ekassos in #1821

Full Changelog: v1385+srv708.web522...v1394+srv716.web524