0.1.50Faster Cube engines, and change anything by chatting

11 November 2026

Our biggest update yet. Both engines got faster, you can now change any card just by asking, and the whole app has been rebuilt to be simpler: a new first run, settings in six plain pages, and a turtle that keeps you company while things load.

Cube 2B and Cube 4B are faster

GreenCube now runs a newer version of llama.cpp (b12211), the engine room under the AI. The same Cube engines now read your message much faster, so answers start sooner, especially when you send a long message or a file.

Reading your message: Cube 2B up to 2.3× faster, Cube 4B 58% faster (measured on our test laptop).

“Less text please, I have to read it out in class” (in Spanish): the same slides, rewritten shorter. Cube 2B, from our test robot.

What's new

• Change any card by chatting. Slides, quizzes and flashcards change when you ask: add or remove a slide, swap two, "make it 10", "less text", or translate it. It works in your language.
• Quizzes in the chat. Answer a quiz right in the chat and GreenCube writes your score and the right answers. Sums are checked by code, not guessed.
• Better slides. Every deck gets a cover page and a proper ending, the junk slides are gone, and look changes like "darker" or "more colourful" apply at once.
• Engines have names. Gemma 4 E2B is now Cube 2B and Gemma 4 E4B is Cube 4B; the bigger ones are Cube 12B, 26B and 31B. Same engines, all built on Google's Gemma.
• A new first run. Sign in, pick your engine on a simple ladder made for your PC, follow an honest download bar with the real speed and time left, then try one of three first things.
• Settings in six plain pages: You, Engine, What it makes, Your computer, Account and Help.
• An update welcome: a gift to unwrap, a note from Hector, your water carried over and three quick tips.
• The water counter grows through 11 levels, from an espresso cup to a swimming pool. From this update on it counts about 5 mL of data-centre water per answer; what you saved before stays.
• Problems in plain words, with one button. Short on memory? "Show me what's using memory" opens Task Manager. And if the app ever crashes, a crash screen saves the logs.
• A new reading font (Literata), and the turtle: it sleeps while an engine loads and paints while a picture is drawn.
• All of it in 19 languages, and when the update arrives, a "what's new" card shows you the highlights in your language.

What we fixed

• The question you asked stays in view while the answer is written.
• The chat works while a card is open.
• A round button jumps to the latest message.
• A security update for a zip library the app uses.

Is it really better?

We ran 25 conversations the new version had never seen on both versions and compared the answers blind. The new version won 14, lost 7, and 4 were ties.