> For the complete documentation index, see [llms.txt](https://docs.beye.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.beye.ai/release-notes/september-30-2026.md).

# September 30, 2026

### September 30, 2026

Claude Sonnet 5.5 and GPT 6.1 Sol join the model picker, and Fable 5 and 5.1 have been withdrawn. You can also now see everything the assistant has remembered about a dataverse and why it chose to remember it, and tool results, code blocks, and message details are all easier to read.

#### New

**Model updates**

Claude Sonnet 5.5 and GPT 6.1 Sol are now available in the model picker.

Fable 5 and Fable 5.1 have been withdrawn and can no longer be selected. Both require a 30 day data retention period from the provider, which is incompatible with the zero data retention arrangement Beye runs under, so they have been removed rather than kept available on weaker terms.

If either was your saved choice, you will be asked to pick a different model the next time you send a message. Conversations you have already had with them are unaffected and still show which model produced each answer.

The model used when you have not chosen one is now Claude Opus 5.5.

**Memory activity log**

The Memories settings page now has an Activity view showing how the assistant's knowledge of a dataverse has changed over time.

Every time the assistant saves, updates, or removes something it remembers, the change is recorded along with the reason it gave for making it, so a memory you did not expect can be traced back to the moment it was created.

Workspace owners and admins see all activity for a dataverse, both shared and personal. Other members with access see their own.

Attempts that were turned down because the 50 memory limit had been reached are now recorded as well, instead of passing without a trace.

**Clearer tool results**

Results the assistant gets back from its tools now appear as a compact tree you can expand and collapse, rather than a block of raw text.

Click any value to copy it. You can also copy a whole section, or the entire result at once.

#### Improved

**Code in chat is highlighted**

Code blocks in the assistant's replies are now syntax highlighted, covering SQL, Python, JSON, JavaScript, TypeScript, shell, and YAML. Other languages render as plain text in the same theme rather than losing their formatting.

Highlighting follows light and dark mode.

**Message details**

Each message now carries its details in a footer inside the bubble: which model answered, how long ago, and the controls to save or share it.

Opening the footer shows input and output sizes, how long the answer took, the reasoning effort used, and the message id.

Copy buttons across the experience no longer raise a notification. The button itself confirms, which keeps the screen quiet when you are copying several things in a row.

**Readable queries**

Queries that an analysis ran are now laid out across multiple lines in the Queries tab. Previously each one arrived as a single long line, which became unreadable as soon as a query selected more than a few columns.