Google didn't just show off new models at I/O 2026, it announced a full-stack shift toward AI that acts, not just answers. Gemini 3.5, Gemini Spark, smart glasses, a new OS approach, and the biggest Search upgrade in 30 years. Here's everything that dropped and why it matters.
OpenAI didn't just release a new voice model today, it released three. Together, they move realtime audio from simple Q&A toward voice interfaces that can reason, translate, and act. Here's everything announced and why it changes the game for developers.
Anthropic's developer event wasn't about a new model. It was about shipping the scaffolding that makes long-running, self-improving AI agents actually work in production. Here's everything revealed and why it matters.
NVIDIA's biggest GTC yet wasn't really about GPUs. It was about the full stack being assembled to make autonomous AI agents production-ready. Here's what that means and why it matters.
A 0.9B model from Zhipu AI just posted state-of-the-art document OCR scores. I read the paper to understand the architecture, the training tricks, and why this is a bigger deal than the benchmarks suggest.
OpenAI dropped GPT-5.4 on March 5, 2026, and it's not just a bump in the version number. This is the first general-purpose AI that can operate your computer, reason through a million tokens of context, and outperform office workers at their own jobs 83% of the time.
Most AI benchmarks test what models know. The Bullshit Benchmark V2 tests something more important, whether they know when a question deserves no answer at all.
Seedance 2.0 is everywhere this week. I read the 1.0 technical report to understand how these models are built - and how they compare to LLMs.