Following yesterday’s Sora Turbo announcement, we’ve finally arrived at the fourth installment of the year-end AI showcase.
The 20-minute presentation was hosted by OpenAI’s Chief Product Officer Kevin Weil, along with Lee Byron and Alexi Christakis.
Specifically, ChatGPT Canvas introduced three major updates:
- Integration of Canvas functionality into ChatGPT’s core model
- Support for direct Python code execution within Canvas
- Introduction of Canvas features to custom GPTs
Following standard practice, OpenAI demonstrated practical applications of these new features.
For instance, Canvas can create a Christmas story about elves, edit titles, polish documents, check grammar, and even add appropriate emojis.
After the presentation, OpenAI CEO Sam Altman emphasized this point on the X platform: “Now, all ChatGPT users can access Canvas functionality and execute code! More importantly, it still maintains its ability to add emojis to your text.”
The team proceeded to showcase more advanced ChatGPT Canvas capabilities.
As a physics enthusiast, the host uploaded a draft paper titled “Santa’s Sleigh: Exploring Dark Energy in Reindeer Propulsion” and asked ChatGPT to evaluate it from a physics professor’s perspective.
As expected, ChatGPT quickly provided professional feedback, making the paper ready for professor review after minor adjustments.
The second major update to ChatGPT Canvas is Python code execution support.
When code is copied and pasted into ChatGPT, it automatically recognizes and switches to code editing mode, assisting with debugging and error detection. Users can run code directly in the interface to quickly identify any issues.
For programmers encountering challenging bugs, this feature offers a convenient solution.
According to official information, OpenAI has integrated a WebAssembly-based Python simulator into Canvas, enabling it to load most Python libraries for immediate code execution.
During the demonstration, when the host requested ChatGPT to create a Christmas-themed Sankey diagram showing Santa’s toy production or distribution flow, the final result fell short of expectations.
However, on a positive note, at least Altman didn’t try to pass off a recorded video as live.
The third significant update is the introduction of Canvas functionality to the GPTs ecosystem.
Imagine writing to Santa Claus this Christmas, with a wish list including a new bicycle, an H100 GPU, and even some “dark matter,” but Santa is too busy to reply. What then?
Using a Santa-themed GPT (Santa Letter Drafter) with the new Canvas functionality, the GPT can perfectly mimic Santa’s tone in reply letters.
Notably, Canvas functionality is disabled by default for existing GPTs but enabled by default for new ones. To enable Canvas in existing GPTs, users simply need to check the relevant option in the configuration interface.
The presentation concluded with a Christmas joke: “How does Santa take pictures? With a North Polaroid camera.”
Looking at the updated features, ChatGPT Canvas appears to have drawn inspiration from products like Claude Artifacts or Cursor, while offering more refined functionality that somewhat surpasses its predecessors.
I also had it write a story about Cinderella and the Seven Gourd Babies.
It can even create and run a working airplane battle game.
OpenAI employee Karina Nguyen also shared some practical applications on the X platform. For example, having ChatGPT explain mathematical concepts and write code to generate charts, using visual aids for more intuitive understanding.
Users can search API documentation, guide the model to write and execute code to obtain desired charts, or easily create personalized tools or games through ChatGPT’s Canvas.
The current ChatGPT Canvas functionality is steadily progressing toward Karina Nguyen’s vision:
“My ideal AGI interface would be a blank canvas that evolves and adjusts according to human preferences. It would innovatively interact with humans, constantly exploring new ways of communication, fundamentally changing our relationship with AI and the entire internet.”
Among these developments, AI code generation has emerged as one of the most valuable applications.
Currently, AI code usage is approaching the critical 50% threshold. This transformation is less about efficiency improvement and more about fundamentally restructuring programmers’ creative thinking and workflow.
Market data strongly supports this development trend.
Public data shows that by 2027, the global software development market size is expected to reach $1.039 billion, with a compound annual growth rate of 22.54%. The success of Cursor reflects the great potential of AI programming tools.
As an AI programming assistant, Cursor has attracted over 40,000 users, with its annualized recurring revenue skyrocketing from 1 million in 2023 to 65 million, an impressive growth rate of 6,400%.
Just before OpenAI’s livestream event, Devin, the world’s first AI programmer, officially launched with a starting price of $500 per month. It integrates directly into our workflow, supporting use in Slack, GitHub, and even private integrated development environments (IDE beta).
In the era of large language models, AI programming competition isn’t about code output quantity but about who can create the most complete development experience loop. The real competition isn’t about who enters the market first, but who better understands developers’ pain points.
From current observations, OpenAI appears to be making a precise serve to win this crucial point.