🚀 OpenAI launches GPT-5.3 Codex with record benchmarks 🚀
New coding model sets industry highs while running 25% faster

OpenAI dropped GPT-5.3 Codex, its most capable coding model to date, achieving new records on SWE-Bench Pro (57%) and Terminal-Bench 2.0 (77%). The model runs 25% faster than previous versions while using fewer tokens. Alongside the model, OpenAI unveiled OpenAI Frontier, a new enterprise platform that helps organizations build, deploy, and manage AI agents with shared context, onboarding, and clear permissions to boost efficiency in real work environments.

💰 Alphabet hits $400B annual revenue milestone 💰
Search giant becomes first to cross historic revenue milestone

Alphabet reported $400 billion in annual revenue with AI powering growth across all business lines. Services generated $350B, Cloud reached $70B run rate, and YouTube alone hit $60B (surpassing Netflix's entire business). CEO Sundar Pichai highlighted Gemini 3 as a major milestone, with the model now processing over 10 billion tokens per minute and the Gemini App surpassing 750 million monthly users. The company plans to spend up to $185 billion on AI infrastructure in 2026, roughly double its 2025 spending.

🔧 Anthropic's Claude builds working C compiler 🔧
Parallel AI agents spent $20K to create production-ready compiler

Anthropic demonstrated how multiple Claude instances working in parallel successfully built a Rust-based C compiler capable of compiling the Linux 6.9 kernel. The project consumed $20,000 across 2,000 sessions, exploring autonomous LLM capabilities through stress tests. While functional for production use, the compiler has limitations including inefficient code generation and reliance on GCC for certain features, highlighting both the potential and current boundaries of AI-driven software development.

INTERESTING TO KNOW

🎥 Meta launches standalone Vibes AI app 🎥

Meta is rolling out a standalone app for Vibes AI, its video generation platform that was previously only accessible through the main Meta app. The move signals Meta's commitment to competing directly with OpenAI's Sora and Google's Veo in the AI video space, giving users a dedicated interface for creating and sharing AI-generated video content. The standalone app aims to simplify access and encourage broader adoption of Meta's video AI capabilities.

🤖 Perplexity launches Model Council 🤖

Perplexity introduced Model Council, a feature that runs user queries across three leading AI models simultaneously (such as Claude Opus 4.5, GPT-5.2, and Gemini 3 Pro). A synthesizer model then reviews all three outputs, resolves conflicts, and delivers one comprehensive answer. The feature helps users confirm and compare the validity of answers generated across frontier models, addressing concerns about AI hallucinations and accuracy in critical use cases.