Model releases, launches, and the odd bit of chaos — the stuff we've been tracking while building Otomas. Newest first.
Google pushed Gemini Omni straight to consumer desktops with a free promo — generate, edit, and remix AI video without touching a professional tool. Ten free videos in, most people finally got what the fuss was about.
Read the sourceGPT-5.6 shipped with programmable tool calling and genuine multi-agent orchestration baked into the Responses API, plus a full-duplex voice mode that finally stops talking over you.
Read the sourceGoogle's keynote leaned hard into voice this year — two new Gemini Live models nicknamed Capybara and Nitrogen — alongside a first look at the Gemini Omni video editor and the unified Aluminium OS heading to laptops.
Read the sourceGPT-5.5 quietly became the model behind the default ChatGPT experience, and the jump in everyday accuracy was bigger than the version bump let on.
Read the sourceOpenAI gave GPT-5.4 the ability to actually use a computer — clicking, typing, navigating — built in rather than bolted on, along with a faster way for the model to search its own tools.
Read the sourceAnthropic's 2026 Agentic Coding Trends Report made it official: teams of coordinated agents were now shipping, debugging, and testing real features over multi-day stretches with barely any hand-holding.
Read the sourceTrump signed an executive order tying AI directly to national defense and infrastructure. The comparisons to the original Manhattan Project weren't subtle, and they weren't meant to be.
Read the sourceAnthropic began sunsetting Claude 3.7 Sonnet and nudging developers toward the newer model family — proof that even flagship models have a shelf life now.
Read the sourceGPT-5 arrived as one system instead of two — a fast model and a slower "thinking" model, with a router deciding which one you actually needed for a given question.
Read the sourcexAI's Grok Imagine dropped with an image generator attached, and the thing everyone actually talked about was its unfiltered "spicy mode," not the underlying tech.
Read the sourceGoogle's newest generation tools made it trivially easy to fake footage of riots, conflict, and election fraud — and the backlash arrived about as fast as the tools did.
Read the sourceVeo 3 and Imagen 4 headlined Google I/O, Gemini got wired deeper into Chrome, and Gemini 2.5 Pro brought "thinking" models to people who'd never heard the term before.
Read the sourceClaude Sonnet 4 landed as Anthropic's answer to "good at coding and good at everything else," extended thinking included, no separate model required.
Read the sourceApple Intelligence finally went properly global with iOS 18.4 — priority notifications, a new Sketch style in Image Playground, and support for a lot more languages than launch day had.
Read the sourceOver a thousand people from a hundred-plus countries sat down in Paris, co-hosted by France and India, to try to actually agree on what responsible AI governance looks like.
Read the sourceElon Musk put a $97.4 billion offer on the table to take back control of OpenAI, reopening every legal and structural argument that had been simmering since he left.
Read the sourceMistral, Europe's answer to the big US labs, put Le Chat on iOS and Android — proof the assistant race wasn't only being run out of Silicon Valley.
Read the sourceChatGPT Deep Research shipped with the ability to run genuinely long research tasks on its own — the kind that used to eat up a person's entire month.
Read the sourceTrump announced Stargate, a $500 billion bet from OpenAI, SoftBank, Oracle, and MGX to build AI supercomputers on US soil. Big number, bigger ambition.
Read the sourceDeepSeek's R1 kicked off the year by matching frontier models at a fraction of the training cost. It briefly outranked ChatGPT on the App Store and knocked billions off Nvidia's valuation in a single day.
Read the sourceWe drop the same updates in the Discord the day they happen, and walk through the bigger ones on YouTube.