The mystery is solved... and the answer is 40x cheaper than Claude

Overview

This episode dives into the rise of Ox Alpha, an anonymous AI model that took the developer world by storm before being revealed as Zhipu's GLM-5.3-Flash. The discussion explores its technical capabilities, the mystery surrounding its origins, its impact on Silicon Valley, and its potential applications, including a humorous experiment with a horse-themed app.

Notable Quotes

- Nobody actually cares about privacy when the magic box is free and works this well. – On developers' willingness to trade privacy for functionality.

- It wrote the app in HTML, CSS, and vanilla JavaScript and wrote code that looks like a legacy codebase of its own. – On Ox Alpha's quirky coding style.

- The entire stealth run was served on just 100,000 Chinese-made chips, which I'm told really hurt Jensen's feelings. – On the disruptive implications of Ox Alpha's hardware efficiency.

🐂 The Rise of Ox Alpha

- Ox Alpha, an anonymous AI model, served 42 trillion tokens in just six days, becoming the most popular model on OpenRouter.

- It features a million-token context window, video input capabilities, and advanced coding skills.

- Speculation about its origins ranged from Google to Chinese labs, with its stack traces and tokenizer fingerprint eventually linking it to Zhipu's GLM series.

🌏 Made in China: Disrupting Silicon Valley

- Ox Alpha was revealed to be GLM-5.3-Flash, a multimodal model with 320 billion parameters, developed by Zhipu.

- Its stealth launch relied on just 100,000 Chinese-made chips, showcasing China's growing influence in AI hardware and software.

- The model's pricing—40x cheaper than Claude—further disrupted the market, though it came with some performance quirks like slower speeds and occasional doom loops.

💻 Developers' Love Affair with Ox Alpha

- Despite privacy concerns, developers eagerly used Ox Alpha for tasks like bypassing rate limits and handling proprietary code.

- Viral hype included exaggerated benchmark claims, though its actual performance still impressed under scrutiny.

- The model's free availability during its stealth phase drove massive adoption, accounting for nearly a third of OpenRouter's weekly traffic.

🐴 Testing Ox Alpha on a Horse-Themed App

- The host tested Ox Alpha by modernizing a legacy Angular.js app called Horse.

- The model demonstrated impressive vision capabilities, detecting CSS bugs, creating animations, and even analyzing video content to generate relevant titles and descriptions.

- Its humorous touches and creative solutions highlighted its potential for quirky, user-centric applications.

🔍 EXA: A Complementary AI Tool

- EXA, a search engine for AI agents, was introduced as a powerful tool for accessing structured data from diverse sources like research papers, financial filings, and company information.

- The host integrated EXA into a stock market gambling app, showcasing its ability to convert SEC filings into structured JSON and cite sources for transparency.

- EXA's compatibility with popular LLMs and its ability to handle complex queries make it a valuable resource for developers.