## Assistants
### New Features
- Introduces new Web Search option in the Tools section for Web
Search-capable models. These include:
    - gpt-4o-mini
    - gpt-4o
    - gpt-4.1-mini
    - gpt-4.1
    - o4-mini
    - o3
    - gpt-5 with reasoning levels `low`, `medium` and `high`
    - gpt-5-mini with reasoning levels `low`, `medium` and `high`
### Known Issues
- Not all Web Search-capable models are currently available on PingPong.
### Notes
- Classic Assistants do not support Web Search capabilities. To use Web
Search, create a Next-Gen Assistant.
- `minimal` reasoning effort does not support Web Search capabilities.
To use Web Search, select a higher reasoning effort level.

## Threads
### New Features
- Introduces support for the full Reasoning feature set of Reasoning
models.
- Adds support for preserving Reasoning summaries. 
- Adds support for saving encrypted Reasoning tokens to be passed as
input in future conversation rounds. Together with Reasoning summaries,
the Assistant can now refer to its previous train of thought when
crafting its next response.
### Known Issues
- There is no UI indicator when the Assistant uses File Search, Web
Search, or Reasoning. As the Assistant takes multiple turns reasoning
and making Tool calls, the Thread UI will show no progress besides a
spinner. Depending on the reasoning configuration and model selection,
Time to First Token (TTFT) may be upwards of 30 seconds.
- Only Web Search result annotations within the assistant's response are
displayed in the Thread UI. Users can't view which search queries the
Assistant made, or the entire list of search results the Assistant
considered to craft its response.
- Reasoning summaries are not streamed or displayed in the Thread UI.
### Resolved Issues
- Fixed: Previous Assistant messages passed as input in future
conversation rounds do not contain the Message ID assigned by OpenAI,
which may cause the API to disregard the `assistant` role of certain
Assistant responses.