scratch's first true artificial intelligence model. out now. (note: i temporary turned the AI off cuz i'm adding something in a private version right now! will be back soon) - - - - - - - - - - - - - - - - - - - - - - - - - - - - please read everything below: 1. omnı is still in early development phases. as you can tell, the UI is not the greatest, the variables struggle with longer answers, and the entire project is plagued with many problems. it is still in development but i would extremely appreciate feedback! 2. if you would like to help omnı, please remix the project and try to make some modifications to the UI. i will be looking through remixes. do note that the AI model will not work in remixes but if it is good, i will replace it with my current project. 3. response times may be longer than expected, especially with the advance model. there may also be outages from providers at times. 4. it is possible for other users to decode and read your messages, so be aware of that. do not enter private information to the AI.
just gonna be some nerdy yap in here! read only if you would like. important info is in the instructions section. - - - - - - - - - - - - - - - - - - - - - - - - - - - - omnı supports two different models inside. omnı-flash, and omnı-advance. both serve different purposes and different architectures, which i will be going over below. omnı-flash = further optimized for speed, based on DiffusionGemma, a 26B A4B MoE model based on Gemma 4 architecture, using block-autoregressive multi-canvas sampling to generate outputs must faster. omnı-advance = custom finetuned version based on Step 3.7 Flash, a 198B A11B parameter MoE LLM. omnı models are hosted on hosted securely on NVIDIA GPUs. enjoy :)