OpenAI’s new model, o1-preview, has shown impressive capabilities in the realm of coding by passing a series of rigorous tests designed to challenge its programming prowess. Released just four months after the GPT-4o model, o1-preview is part of the ongoing advancements in AI that OpenAI continues to deliver.
Dubbed “Strawberry” during its development, o1-preview is now available to ChatGPT Plus subscribers and brings enhanced reasoning capabilities, breaking down complex prompts into manageable steps. This feature is showcased through a reasoning summary visible before each response, adding a layer of transparency to how the model processes information.
In practical tests, o1-preview successfully wrote a WordPress plugin, managing to separate duplicate entries in a list without placing them next to each other—a task that previously stumped many models. Additionally, it adeptly rewrote a flawed string function, pinpointed the real issue behind a misleading error in WordPress API usage, and skillfully integrated knowledge across different coding domains to write a functional script combining AppleScript, the Chrome DOM, and Keyboard Maestro.
Despite its capabilities, the model’s responses tend to be verbose, providing detailed explanations and reasoning steps that, while informative, may overwhelm users seeking more concise information. This characteristic illustrates the model’s thorough understanding but also highlights the need for customization in output verbosity to suit different user preferences.
The performance of o1-preview not only enhances the functionalities of AI in coding tasks but also promises further integration with advanced features like file analysis and web access, expanding its utility in real-world applications. For those interested in the specifics of these tests and the model’s performance, you can find more details here.