On Dec 31, 2025, will a widely available AI model be able to write a sophisticated 2000 line program?

Duplicate of this question. Same resolution criteria but resolution date is 12-31-2025.

For positive resolution, the AI model must be widely available, meaning: I personally can access it via a free trial, modest subscription (of say <$50) or an API call (again <$50)

Examples of models that are (as of 12-28-2024) publicly available: gpt-o1, gemini-2-thinking, claude-sonnet-3.5, Deepseek v3.

Examples of models that are not (as of 12-28-2024) widely available: gpt-o1-pro, Devin, gpt-o3.

I will make my best effort to determine if there is a model that would allow a positive resolution (including writing the 3000 word prompt and running the resulting program), but others are also allowed to submit solutions, assuming they follow the correct format. It is allowed to prompt-engineer, as long as there isn't any obvious code golf going on. The prompt should look like a English language description (with some technical details included).

An example of what I mean by a "sophisticated program" would be: "write a gatcha game similar to Genshin Impact where all of the art is generated using AI models"

Specific technical requirements:

The program

• is at least 2000 lines of code

• has multiple user interfaces

• Involves storing/retrieving/modifying multiple types of objects from a database

• has multiple concurrent threads

• the flow of the program depends in some significant way on the count/states of different objects in the database

• The program depends on combining the output of multiple cutting edge AI models

• At least one of the models is less than 30 days old (meaning code that could run the model did not exist 30 days earlier)

The Challenge:

• The AI will be given a design document in plain English of no more than 3000 words.

  • The document may include suggestions such as "use library X" or "use model Y"

  • the document may include a flow diagram (in text) of how the program should operate.

• The AI must then write the entire program without human intervention

  • it is allowed to use methods such as Code Interpreter or Web Browsing, the only thing it cannot do is receive human assistance

• The AI must produce a complete program which runs correct and meets all of the requirements in the design document

• The AI can have as many "tries" as it wants, provided it correctly identifies which is the successful program (a human is not allowed to pick the best program for the computer)

• Update 2024-28-12 (PST) (AI summary of creator comment): - Specific Technical Requirements:

  • The AI can take a >3000-word prompt and return as output at least 2000 lines of code satisfying those requirements.

  • The program compiles and runs successfully.

  • The GUI is actually usable by a human being.

  • The creator reserves the right to add requirements that are implied but not stated and will be extremely generous in the AI's favor if the resolution is even remotely close.

• Update 2024-28-12 (PST) (AI summary of creator comment): Update from creator

  • The AI can take a 3000-word prompt and return as output at least 500 lines of code satisfying those requirements.