What's new in NextDocs, summer 2026
We have been quiet on this blog since the Shyne announcement, and quiet reads as absent. It was not. This is the catch-up: everything that shipped in NextDocs from v1.10.3 in June through the September model changes, in one place.
Bring your own fonts (v1.11.0)
The most requested feature of the year. You can now upload font files (WOFF2, WOFF, TTF and OTF) or import any font from Google Fonts, assign them as theme fonts, and use them anywhere text appears. Exports respect them: uploaded fonts are embedded in PDF, PNG and HTML exports, so what you see is what the reader gets, not a fallback.
v1.11.1 added font favourites and upload from inside the editor, so a brand font is a click away instead of a settings trip.
Export to Word (DOCX)
Documents now export to Word with headings, lists, tables, images and fonts carried across. If your client wants a file they can redline, you no longer need to rebuild it. The full post is here, including what carries over and what to expect from equations and code blocks.
Google Docs export also got sharper: colours and highlights now survive the trip.
Long documents that finish (v1.11.1)
Very long decks and reports were the place where generation used to fail most often, usually near the end. That is fixed: long runs complete reliably, and if a run is interrupted, NextDocs keeps the pages it finished rather than throwing the whole document away. The how and why is in its own post.
Richer text on the canvas
Lists, block quotes, code, links and underline are now first-class on the canvas, with pixel-accurate text editing. The editor also got calmer: fewer accidental moves, touch multi-select, and zoom while cropping an image.
New models on every lane
Three changes since July:
- The fast lane runs on GPT-5.6 Luna, hosted on Azure AI Foundry.
- The quality lane moved to Gemini 3.8 Flash on 3 September, after it passed our long-document tests.
- The model picker now offers Gemini 3 Flash and 3.8 Flash, Gemini 3.1 Pro Preview, GPT-5.5, Claude Opus 4.8, Claude Sonnet 4.6, Kimi K2.6, GLM 5.3 and Qwen 3.7 Max.
If you want to understand what those names mean and which to pick, read the September model wave, explained.
AI Memory
NextDocs now remembers your role, audience, tone and standing preferences across chats. Set them once, or just tell the AI to remember something, and every new document starts from context instead of zero. Read the announcement.
Smaller things you will notice
- Help reaches a person. The Help & Feedback button now files a ticket that lands with our team within minutes.
- Trash. Deleted documents go to a trash you can restore from.
- PPTX fixes. Colours and text sizing in PowerPoint exports match the canvas.
- Longer instructions. Custom instructions can be far longer than before, which matters if you paste a style guide.
What is new at Shyne
Shyne is where our new kinds of capability land, and it has had a busy month too: export to Google Slides shipped on 9 September, publishing goes to Instagram, Facebook, LinkedIn, TikTok, Threads and YouTube from the canvas, the desktop app runs on macOS and Windows and can use your own Claude Code or Codex, and the Shyne blog is live. If you are curious what a shared canvas with an agent on it feels like, that is the place to look.
What is next for NextDocs
More of the same, on purpose: model upgrades when a model earns its place, export fidelity, reliability, and the fixes you report. That is the job NextDocs does, and it keeps getting better at it.
Open NextDocs and try the new fonts on something you already have.
The NextDocs Team
