Model Updates and AI Improvements
Today’s update refreshes the system prompt model lists with the following additions and changes:
Removed models that are no longer supported by their vendors (e.g. older Gemini and Mistral).
Included more modern and alternative models that perform well at their tasks (e.g. GLM, MiMo and Aion)
Existing Chats using removed models from the old system collection will keep their settings.
As always, you can find previous system model collections in the archive.
AI/Prompting
Replaced the “Try for a spin” with a search modal for easier access than a menu. This also fixes the issue of the site hanging when listing models from Featherless.
You can now pick any one-off model for a Chat thread, without the need for a model collection.
Fixed the pov prompt functions not working with equality and contains checks - especially when comparing POVs with characters. This should improve the output of the default scene beat prompt when dealing with multi-POV stories.
Disabled auto-cache-write requests for GPT 5.6/6/6.1 models when used via OpenRouter. Previously, cache writes were offered for free, but now they’d cost 25% extra.
Updated the AI vendor icons for OpenRouter, Google AI Studio and Featherless, since they’ve all changed their branding.
Other Changes
- We gave the settings modal a little visual polish. Same stuff as before, just easier to find.
- Added an option to show plain loading screen messages instead of the playful ones.
- Various other fixes and improvements.