The AI Text Editor now supports streaming responses. Instead of waiting for the entire response to be generated, you can see the AI's output appear in real time as it writes, making the editing experience feel much more responsive.
Updated the available AI models across the board. GPT-4o and GPT-4o mini have been replaced with GPT-5.4 and GPT-5.4 mini. Claude 3.5 Haiku is now Claude 4.5 Haiku, and Claude 4 Sonnet has been upgraded to Claude Sonnet 4.6. Qwen 2.5 72B has been replaced with the much more capable Qwen 3 235B. DeepSeek-V3 and DeepSeek-R1 have been consolidated into DeepSeek-V3.1. Additionally, Gemini 1.5 Pro, o4-mini, and Kimi K2 have been removed.
You can now customize the Study Note colors.
Streaming Study Materials is now also supported inside the Concept Maps.
Study Materials like Flashcards or Quizzes are now generated via streaming which makes them load faster
Improved the Flashcard and Quiz pages for a better learning experience. You can now sort your flashcards or select them individually to apply various transformations. (I use this feature to create short stories from my German word flashcards). For quizzes, you can now ensure they are constantly shuffled to ensure the learners don’t memorize the quiz answers by their positions.
Updated a lot of pages to work much better on mobile screens. This way, you can now memorize your flashcards or take your quizzes whenever you have your mobile phone with you.
You can now chat with any selected text using the pop-up chat window. This way you can get answers to small questions you have without having to initiate a whole conversation with AI using the Chat Sidebar.
Fixed several performance issues on the PDF pages. As a result, large PDFs should load much faster (10x increase) and the whole experience should be much smoother.
When using LLMs, context is everything. It is a known fact that a large context deteriorates the LLM performance, and that’s why you probably never want to upload a huge PDF and ask it to summarize it for you. You will likely not get an accurate or even a decent answer. You can now use the range controls in Heuristica to restrict the context to the parts that you care about, so you can get better answers for your questions.