DHH's New Programming Model: DeepSeek $23 Completed, Fable Around $550

According to Perceive Beating monitoring, Ruby on Rails creator DHH conducted a single Python to Rust conversion task test across multiple models. The DeepSeek Pro V4 Max was successfully completed after approximately 2.5 hours, with a total token cost of $23. Grok 4.6 cost $55 and took about 1.5 hours, while GPT-5.6 Sol cost $43; both DeepSeek V4 Flash and GPT-5.6 Luna were unable to complete.

In the initial round led by Fable, DHH calculated a cost of approximately $550 and a duration of about 45 minutes. It involved rewriting TerminalTextEffects from Python to Rust, consuming around 11 million tokens. The startup time decreased from 87ms to 2ms, and the rendering speed increased by 9.6 times.

However, there is a caveat to the comparison. Fable initially drafted a 483-line implementation plan, which was reviewed by Codex xhigh, and all other models directly adopted this plan. Therefore, this is more like comparing the cost and efficiency of different models executing the same plan.

The project has been open-sourced. The repository includes 354 consistency tests that compare the Rust version and the original Python version frame by frame and byte by byte. The final artifact can be compiled into a static single file of around 3.3MB.