Google must be cooking up something big...

Key Concepts

Gemini 2.5 Pro, LM Arena, LiveBench, OpenAI, Public Benefit Corporation, Windsurf, VS Code fork, 3JS, Vision Prompts, Full-Stack Application, PostgreSQL Database, Savala, Google Kubernetes Engine, Cloudflare, Database Studio.

Gemini 2.5 Pro and AI Model Rankings

Google released Gemini 2.5 Pro, which is currently ranked number one in all LM arenas, making it arguably the best coding AI model. This release is surprising because it precedes Google IO, their annual developer conference, where they typically announce major tech advancements. The speaker speculates that Google might be developing something even bigger, such as Gemini 3 or Gemini 2.5 Ultra, given their focus on using AI to develop AI.

• LM Arena: A platform for blind taste tests where users evaluate LLMs based on their responses. Gemini 2.5 Pro is currently dominating this arena.
• LiveBench: A more scientific benchmark that uses contamination-free questions to evaluate LLMs. OpenAI models still outperform Gemini on LiveBench.
• Coding Performance: While Gemini 2.5 Pro excels in coding benchmarks, it reportedly falls behind its previous versions on other benchmarks.

OpenAI's Shift to a Public Benefit Corporation

OpenAI is abandoning its plans to become a for-profit company after facing criticism, particularly from Elon Musk. Instead, they are transitioning to an uncapped profit public benefit corporation (PBC).

• Public Benefit Corporation (PBC): A type of corporation that is legally obligated to consider the interests of society and the environment in addition to profit.
• Uncapped Profit: The PBC structure allows OpenAI to generate unlimited profits while maintaining the appearance of prioritizing public benefit.
• Nonprofit Oversight: The original nonprofit will continue to oversee the PBC and act as a shareholder.
• Comparison to Other AI Companies: Anthropic and XAI are also structured as public benefit corporations.

OpenAI's Acquisition of Windsurf

OpenAI is acquiring Windsurf, a VS Code fork, for $3 billion. This acquisition is surprising given OpenAI's claims that their AI is among the top 50 programmers in the world. The speaker questions why OpenAI would spend such a large sum on a VS Code fork when they could potentially build one using their own AI.

• Windsurf: A fork of the VS Code editor.
• VS Code Fork: A separate version of the VS Code editor that has been modified or customized.

Hands-on Testing of Gemini 2.5 Pro

The speaker tests Gemini 2.5 Pro's coding capabilities, including generating Svelte 5 code and building a game from scratch using 3JS.

• Svelte 5: A JavaScript framework for building user interfaces.
• 3JS: A JavaScript library for creating 3D graphics in a web browser.
• Svelte 5 Code Generation: Gemini 2.5 Pro generated syntactically correct Svelte 5 code for a to-do app, but the app did not function correctly.
• 3JS Game Development: The speaker found the results of using Gemini 2.5 Pro to build a 3JS game to be solid but not significantly better than other models.
• Vision Prompts: Gemini 2.5 Pro is particularly strong with vision prompts. The speaker sketched an application on toilet paper and asked Gemini to build a full-stack application with a PostgreSQL database. The resulting application had "pretty solid vibes."

Savala: A Deployment Platform

Savala is presented as a modern successor to Heroku, allowing users to deploy full-stack applications, databases, and static websites using Google Kubernetes Engine and Cloudflare.

• Savala: A deployment platform that simplifies the process of deploying applications.
• Google Kubernetes Engine (GKE): A managed Kubernetes service provided by Google Cloud.
• Cloudflare: A content delivery network (CDN) and security company.
• Database Studio: A feature that allows users to manage their databases directly from Savala's web UI.
• Deployment Process: Users can connect a Git repo or Docker image to Savala, provision resources, and deploy their application.

Conclusion

Gemini 2.5 Pro represents a significant advancement in AI coding capabilities, although its performance varies across different benchmarks. OpenAI's transition to a public benefit corporation and its acquisition of Windsurf raise questions about its long-term strategy. Savala offers a streamlined solution for deploying full-stack applications. The AI landscape continues to evolve rapidly, with new models and platforms emerging regularly.