All patch notes

Patch Notes B0.048 - GPT 6, Opus 5.5, Space Bunny Alpha

Model Changes

Before we get into the specific changes, we wanted to tell you a little more about our process for updating models available in Craft, and why you may notice differences in models over time.

Especially when it comes to open source models like GLM or Deepseek, there are many inference providers who serve them. We use OpenRouter to route between many of these providers. As we find certain providers with better or worse quality, we will switch and pin to providers who meet the best combination of price, output quality, and speed.

These inference providers may change the way they serve the models without notifying us. Providers will often fluctuate in speed, quality, and price, and the reasons are not shared with us. For core models specifically, cost is a big factor in the providers we are able to select as we offer unlimited use of these models. Sometimes, a great provider will all of a sudden increase their prices, leaving us to route to other providers who may not be as good or fast. In these situations, we are always evaluating all of the provider options to make sure we’re serving you the best option we can.

Now for the updates:

  • Added Space Bunny Alpha, an experimental stealth model with image support. Free for a limited time while it is in stealth! It’s quite fast and smart, we recommend you try it and use it as much as you can before it becomes paid.

  • Replaced Opus 5 with Opus 5.5. Opus 5.5 is much more intelligent, faster, and 20% cheaper than the previous Opus 5 model.

  • Added GPT-6 Sol and GPT-6 Luna. The new generation of GPT-6 models are also faster and cheaper than the previous generation. GPT-6 Luna is a core model and is now the default model for new games.

  • Muse Spark models are temporarily unavailable because of a provider issue. If you had one selected, or a game recommended one, those games now run on GPT-6 Luna instead. We’re working on getting Muse Spark back online as soon as possible.

  • GLM 5.3 and Deepseek V4 Flash have had their providers updated. They may not be the fastest providers, but the quality should be a lot better!

  • We’ve heard many reports about Mimo v2.6 Flash getting stuck in reasoning loops. We’re un-retiring Mimo 2.5 as another core option.

  • Retired Mercury 2.5.

Visual Novel Updates

  • You can now start a blank visual novel straight from a project, no game required. Open the new Visual Novels tab in your project settings to create one from a blank page. It uses your project's cover, characters, locations and Game Master for its cast and scenes.
  • You can now set default positions for cast members on a page, so you can place a character once instead of on every page. Individual pages can still move them for custom positions when needed.
  • Video pages can now have an optional second clip made for mobile devices, which plays instead of the desktop clip on narrow screens.
  • You can now preview visual novel pages in mobile view, and set mobile speciifc positions for characters; if mobile positions are not set, it will use the desktop positions as before.
  • You can now upload your own narration audio to any visual novel page.
  • You can now generate expressions and cutouts for your cast inside the visual novel editor, from each character's portrait.
  • Readers can now pick up where they left off on their device: reopening a story offers Continue reading from their last page. Auto play also now stops on the last page instead of jumping straight to the end screen.
  • You can add images to visual novel pages. Place an image from your studio or an upload, such as a prop, a note, or a sign, then you can drag, resize and rotate it, and choose whether it sits in front of or behind the speaker.

CleanShot 2026-09-24 at 13.35.56@2x

  • Craft CLI users can now write, check and publish visual novels from the command line with craft vn, and cloned workspaces now include instructions that teach coding agents how to write them. Update to the latest CLI to use it and create visual novels with outside AI editors.

Improvements and Fixes

  • You can now upload your own videos (MP4 or WebM, up to 25 MB) to your Studio, and upload or pick videos directly in the visual novel editor and in creation flows. Maximum duration for videos in character creation flows has increased to 60 seconds from 30 seconds.

  • When Orbit or the Game Master sets an image on a file, the update card now shows the the image instead of the URL.

  • When a model's provider is temporarily at capacity, you now see a message saying so, with a suggestion to try again or switch models, instead of a generic error.

  • Fixed map pins and areas in larger projects showing a warning icon and "Untitled" even though their linked locations still existed.

  • Long game master and orbit replies should now stream more smoothly.