July 29th, 2026
New!
Improvement
Fix
Since our last roundup we’ve added a new AI model, made Chat way better at knowing which words you’re talking about, and fixed a batch of bugs you reported. Here’s what’s new.
We’ve added Kimi K3 to Write, Draft, and Plugins. Kimi K3 is a new AI model praised for its creative writing chops, sitting at #2 on EQ-Bench’s creative writing leaderboard (second only to Claude Opus 5, which we added last week). It’s apparently quite good at register: ask for hardboiled or lush and it changes its sentence rhythm, rather than just adjectives. Audition it on a scene with a strong voice and let us know how the prose lands.
We improved the way Undo handles Chat-based edits. If Chat had edited your manuscript earlier in a session, pressing Ctrl/Cmd + Z later could roll the document back to before that Chat edit instead of undoing only more recent manual edits. Undo now reverses the most recent change chronologically, the way it should.
We’ve been very focused on improving how Chat handles your instructions.
No more “your manuscript is empty” - It was possible for Chat to write you a useful chapter critique… and then discard it in favor of telling you your manuscript was empty. We fixed that!
Highlights stick to the correct document - When you highlighted text and asked Chat about it, it was possible for Chat to reference a different file in your sidebar—or claim it couldn’t load your chapter at all. We fixed that too, and it should better understand that highlights are associated with your active document.
Accurate word and mention counts - Ask how long a document or project is, or how many times a word appears in a document, and Chat now actually counts instead of estimating.
Pasting editable text into Chat - Pasted text was turning into an uneditable chip instead of dropping into the message box as plain, editable text. We changed that so it remains editable.
Long requests don’t vanish - A big Chat request that ran for many minutes could end in a bare “Load failed” with nothing saved, forcing you to start over. We made long-running requests like that preserve their work.
Comment replies are reliable again - Sometimes @Sudowrite replies in document comments could fail with a generic “I wasn’t able to generate a reply” error. We’ve resolved that issue.
Complete sentences from Brainstorm - Long Brainstorm ideas were being truncated, leaving sentences incomplete in some cases. You’ll get a complete idea from now on.
Worldbuilding trait edits persist - We fixed an issue that caused Worldbuilding traits to overwrite themselves as you typed into them. (Autosave was a little overeager, but it’s tamed now.)
Smart Dictation (on mobile) preserves nicknames - Inside a project, Smart Dictation was “helpfully” expanding a spoken first name or nickname into a character’s full Character Card name. It now writes the name as you said it.
Thanks as always—and please keep the bug reports coming.