Equations now render more reliably and with better performance. This includes fixes for LaTeX rendering and other equation related issues.

You can now log in with your X account.

Claude Sonnet 4 and Claude Opus 4 have replaced the old 3.7 models. These models are all around improvement over the previous ones. They are now available for all users.

GPT-image-1 model has been added and is now available for all users. It is a image generation model that can generate images from text prompts. Feel free to try it out!

A new experimental version of the Gemini flash model has been released. Currently thinking is always enabled but this will be changed in the near future and you will be able to select if you want thinking or not.

You are now able to see the exact details of each AI response. Just hover over the information icon next to the model name at the end of a AI message and you will see details such as token usage, answer time and more.

o4-mini has been added to the models list and replaces the old o3 mini model.

The new GPT-4.1 model and the mini and nano variants have been added to the models list. These models generally surpass the performance of GPT-4o and GPT-4o-mini which have now been archieved.

Refine your search results with the new settings. You can now select the search source to get more relevant results.

You can upload PDF files and other text files to any of the offered models.

You can now adjust the reasoning effort of the o1 and o3 mini models. You can find a new dropdown in the model settings. This allows you to set how long the model thinks about the problem.

You can now upload documents to the chat. The AI will read the document and answer questions about it. Any file can be uploaded but the AI performance can vary between different file types.

Googles new Gemini 2.5 pro model has been added to the models list. It has very low rate limits so prepare for failed requests.

The new Deepseek V3 has been added to the models list. It is a flagship non-reasoning model from the team behind Deepseek R1. Try it out!

You can now include images in your chats. Simply use the new attachment button to add images to the chat. You can ask the AI to describe the image or ask questions about it. The UI is still rough and will be improved shortly. Additionally, other document types will be supported soon.

The chat box has been redesigned and now increases in size as you type more text. Other minor improvements have been made to the user interface.

Mistral small 3.1 has been added to the models list. It is a small model that performs generally better than gpt-4o mini. Try it out!

It is now possible to use the deep search mode with the web search feature. This allows the AI models to read full web pages to provide indepth information. New settings button has been added to fine tune the search settings.

During this week focus has been on some low hanging performance bottlenecks on the backend. The backend improvements do not bring any new features but the website backend is now more ready for larger amount of usage. Almost every request is now at least 40% faster.

The landing page has been redesigned to be more user friendly and to provide more information about the service.

You can now use the web search feature with all models. This feature allows the AI models to search the web for information. Currently a shallow search is performed but deep research mode is coming in the future.

OpenAI o3 mini and o1 models have been added to the models list. They present worlds leading reasoning models and should perform well on any task.

Qwen QwQ has been added to the models list. Qwen QwQ is the cheapest reasoning model available and is closely matched in performace with Deepseek R1.

You can now login with discord. Your discord email has to be verified to use this feature.

You can now select your default model from the profile page. This will be the model that is used when you start a new conversation.

The new claude 3.7 model has been added and supports thinking capabilities. Suited especially for developers.

OxyAI chat is now available in open beta. We are excited to have you here.