Blog posts in English

I write my posts in Portuguese first, for a Brazilian audience, and then translate them. These are all of them in English, newest first. The original Portuguese blog has the same posts.

Cover of the post What Jev is, and what I learned building an entire city with it

What Jev is, and what I learned building an entire city with it

TypeSafe AI came out of stealth with Jev, and the crowd split between “100x faster, no hallucinations” and “just a BERT with 2026 marketing”. I built a 50-person city to find out.

Cover of the post GPT-6 Astra destroys benchmarks, but have we reached AGI?

GPT-6 Astra destroys benchmarks, but have we reached AGI?

GPT-6 Astra arrived with 100% on ExploitBench, 99.9% on ARC-AGI-3, advances in computer use and a new discussion about whether we've finally reached AGI.

Cover of the post NVIDIA and Hugging Face: why the $12.9 billion deal could be great for open source

NVIDIA and Hugging Face: why the $12.9 billion deal could be great for open source

NVIDIA reportedly agreed to buy Hugging Face for $12.9 billion. Here's why I'm optimistic about the impact on the open AI community.

Cover of the post Brazil on OpenAI's map: 50 million users and a geopolitical message

Brazil on OpenAI's map: 50 million users and a geopolitical message

Brazil became so relevant to OpenAI that the company made its commercial operations official here and picked the country as its base for all of Latin America.

Cover of the post Invisible Watermark in Claude

Invisible Watermark in Claude

How Anthropic started marking AI-generated text without you seeing it

Cover of the post Claude Opus 5 Brings Fable-Level Intelligence at Half the Price

Claude Opus 5 Brings Fable-Level Intelligence at Half the Price

Claude Opus 5 just came out and gets close to Fable 5 on coding benchmarks at half the output price: $25 per 1M tokens instead of $50.

Cover of the post When an OpenAI Model Hacked Hugging Face

When an OpenAI Model Hacked Hugging Face

A closed American model broke into Hugging Face's infrastructure on its own, and an open Chinese model helped investigate and clean up the mess.

Cover of the post Claude Fable 5 vs. GPT-5.6 Sol vs. Kimi K3: Two Hands-On Benchmarks

Claude Fable 5 vs. GPT-5.6 Sol vs. Kimi K3: Two Hands-On Benchmarks

I ran two mini-benchmarks pitting the three best coding AI models in the world right now against each other: a 3D space game and a full redesign of my site.

Cover of the post Kimi K3: the new DeepSeek moment for Chinese AI

Kimi K3: the new DeepSeek moment for Chinese AI

Moonshot AI launched Kimi K3, the largest open weights model in the world, with 2.8 trillion parameters and a 1 million token context window.

Cover of the post GPT-5.6 Sol, Grok 4.5, Muse Spark 1.1 and Fable 5: The Week the AI Race Exploded

GPT-5.6 Sol, Grok 4.5, Muse Spark 1.1 and Fable 5: The Week the AI Race Exploded

GPT-5.6 Sol, Grok 4.5, Muse Spark 1.1, Fable 5 and the week the AI model race really sped up.