Summary of OpenAI 12-day series of press conferences

OpenAI officially announced a 12-day series of launch events on December 5, launching an important new product or technology every day. Below are the key highlights and summary of each day.

Day 1: O1 full blood version is online & ChatGPT Pro subscription service

O1 full-blooded version is the strongest inference model launched by OpenAI, with performance improved by 34% and error rate reduced by 35%. It supports multi-modal input and can handle complex tasks of text and images. At the same time, the ChatGPT Pro subscription service is officially launched, and users can get unlimited access to O1 and the professional version of O1 by paying US$200 per month.

• O1 Basic Edition : Suitable for daily use, it improves language understanding, text generation, translation and summarization capabilities, and the response speed is increased by 50%.

• O1 Pro version : For professional users, it supports complex reasoning and professional domain knowledge, and is suitable for handling complex problems such as mathematics, science, and programming.

Day 2: Reinforcement Fine-Tuning

OpenAI introduces reinforcement fine-tuning technology, a new model training method that allows users to create customized expert models with a small amount of training data. It is expected that enhanced fine-tuning will be available to users in the spring of 2025 and is suitable for complex tasks and professional fields (such as law, finance, medical, etc.).

• Enhanced fine-tuning features : Use very few samples to optimize the model, and use reinforcement learning algorithms to continuously improve output quality.

Day 3: Sora official version released

Sora is a video generation tool launched by OpenAI that can generate realistic videos based on text descriptions. Sora supports 1080p resolution, video generation of up to 20 seconds, and provides a variety of editing functions. Open to ChatGPT Plus and Pro users only.

• Highlight function : Supports a variety of video editing functions (such as Remix, Re-cut, Storyboard), and can generate high-resolution, complex scene videos.

Day 4: ChatGPT Canvas is fully open

The ChatGPT Canvas function is officially open, allowing users to collaborate on writing and programming with ChatGPT. Canvas provides users with a shared canvas that supports real-time collaboration and multiple creative forms.

• Highlight features : Supports Python code execution, image recognition, and real-time editing of documents and code.

Day 5: ChatGPT integrated into Apple Intelligence

OpenAI announced a partnership with Apple to integrate ChatGPT into Apple Intelligence . iPhone, iPad and Mac users can directly access ChatGPT’s features through Siri, providing a more powerful smart assistant experience.

• Highlighted features : Siri supports more complex tasks, and Apple’s writing tools and visual intelligence features will introduce support for ChatGPT.

Day 6: OpenAI version “Her” is online

OpenAI adds video input and real-time screen sharing capabilities to ChatGPT’s advanced voice mode. This allows ChatGPT to "see" the user and provide more accurate feedback based on this visual information. This feature is for mobile users and is open to Plus and Pro users.

• Highlight features : video input and screen sharing, supports 50 languages, and enables natural voice interaction.

Day 7: ChatGPT Projects feature released

OpenAI has launched the Projects function, which allows users to integrate related files, chat records and other information into a project for management. This feature integrates Canvas and ChatGPT Search , is suitable for Plus/Pro users, and will be opened to enterprise and education users in the future.

• Highlighted features : Supports multi-step tasks, personalized instructions, file uploads and data management to facilitate project collaboration.

Day 8: ChatGPT Search fully upgraded

The ChatGPT Search function has been fully upgraded on the 8th day. Users can obtain news, sports scores, stock quotes and other information through real-time search. This feature is free and available to users around the world and works with third-party search providers to enhance the accuracy of real-time data.

• Highlight features : Search speed is faster, voice search and map search are added to improve the search experience.

Day 9: Open API for O1 inference model, reducing cost by 60%

OpenAI announced that the O1 inference model API is officially open, reducing the cost by 60%, and adding new functions such as function calling, developer messaging, and image analysis. This technology upgrade not only makes the model faster, it also reduces the price.

• Highlighted features : Supports multiple programming languages ​​(such as Go, Java), introduces WebRTC support, and significantly improves real-time API response speed.

Day 10: Using ChatGPT via phone and WhatsApp

OpenAI has launched a new feature for using ChatGPT via phone and WhatsApp , users can call (1-800-CHAT-GPT) or talk to ChatGPT via WhatsApp. This feature lowers the barrier to use and makes AI services more accessible.

• Highlighted features : voice conversations on the phone and WhatsApp, supports multiple languages, and adapts to different devices (smartphones, elderly phones, landlines, etc.).

Day 11: ChatGPT is deeply integrated with Mac applications

OpenAI announced the deep integration of ChatGPT with Mac applications , allowing users to collaborate with applications (such as Warp and XCode) through copy and paste to improve programming and writing efficiency. This feature is already available on the Mac platform, and Windows users will get it later.

• Highlight features : Seamless collaboration with development tools and document editing tools, supporting real-time code auto-completion and Git submission data analysis.

Day 12: The latest inference model O3 is released

OpenAI released its strongest inference model O3 , which claims to be close to general artificial intelligence (AGI) and surpasses previous GPT-3 and GPT-4 in many fields. O3 has demonstrated excellence in mathematical reasoning and programming competitions and plans to work with the ARC-AGI Foundation to develop the next benchmark.

• Highlights : Excellent performance in mathematics and programming, especially in the American AIME mathematics competition and Codeforces programming competition. O3 provides a clearer reasoning path and accurate answers.

Summarize

This 12-day series of conferences showcased OpenAI’s technological breakthroughs and innovations in multiple fields. From the launch of the powerful inference model O1 to the revolutionary video generation tool Sora, to cooperation with Apple and enhanced voice interaction capabilities, OpenAI is constantly pushing the boundaries of AI technology and improving user experience. Every day's releases bring new features and more advanced tools to users, marking the in-depth development of artificial intelligence applications in multiple fields.