Product Update: September 2026

Here's what went live in September:

• Changes in Excel: See every cell the agent changed, before and after, and undo anything you don't want.

• Sites from chat or email: Describe the Site, ask for changes, tell the agent to publish.

• Multi-agent runs: Bigger tasks now split across focused agents running in parallel.

• Cross-file connectivity: Agents can now work across your open workbooks, sending requests from one file to another.

• Multiplayer chats: Work in the same chat as your colleagues at the same time.

• Site analytics: See who opened your Site, which version they read and how long ago.

• Edit mode in Sites: Edit the text directly, with every change saved to version history.

• Model ML in Slack: Tag @Model ML in any channel or DM and the agent does the work right in the thread.

• Scheduled Site updates: Sites can now refresh themselves on a Schedule, ensuring the data is always up to date.

• HSBC global rollout: HSBC rolled out Model ML globally for research, analysis and drafting in HSBC's own formats.

• Bigdata.com partnership: Premium news, expert interviews, transcripts and sentiment data, now live in Model ML.

• Evals for AI slop: Our harness now uses millions of real-world finance outputs to find and remove AI slop.

• Call transcript review: Our Document Review Agent now validates against call transcripts.

• Agent comments: Leave a comment, send it to the agent, and it resolves it for you.

• Compare outputs for difference: Try it from email, chat, or any surface.

• Fine-tuning experiment: We fine-tuned GLM 5.3 for slide quality checking. It beat GPT-6 Astra and Claude Opus 5 at around 95% less cost.

• Formatting in Sites: Format your Site yourself, from headings, lists and quotes to size, colour and spacing.

Sites from chat or email

You can now build sites through chat and email. Get started by describing the Site you want, then ask your Agent for any changes. When you’re ready to go live, ask it to publish.

Multi-agent runs

Model ML now splits larger tasks between focused agents running in parallel by default. The results are then brought together into a single output. Especially useful for tasks like research and document creation.

Multiplayer chats

You and your colleagues can now work in the same chat at the same time. Simply share a project chat in Collaborate mode and members can ask questions, add instructions and continue the work together. Each message shows who sent it.

Site analytics

Track engagement on every Site you send by selecting 'View analytics' from the deployment menu. Review Site visits, see which versions people opened, and check which named recipients haven't visited yet.

Edit mode in Sites

Turn on Edit mode to change text yourself. Every edit is saved to version history, and the Agent builds on your changes rather than overwriting them.

Model ML in Slack

Hand your agent a task without leaving Slack. Tag @Model ML and it does the work right in the thread, with the same connectors and documents as in the web app. In channels, tag it in each message.

Scheduled Site updates

Your Sites now refresh themselves on a Schedule, set up from chat or the studio. Choose how often it runs, and each refresh will pull in the latest data, keeping the Site up to date whenever someone opens it.

Call transcript review

Your document review agent now checks against more sources. Validate directly against call transcripts, or use the same cross-validation checks you already rely on with Excel documents as the target.

Agent comments

You can now send comments on your Site straight to your agent. It makes the change on the Site itself, then replies to your comment with an update of what it did. Comments also now support emoji reactions and @mentions.

Compare outputs for difference from any surface

Surface every change between two versions of the same deck from email or any chat-based surface.