April 16 - Codex uses your mac in the background, Opus 4.7 release not quite Mythos + 3 interviews

Hey ya’ll, Alex here with your weekly AI news catch up. It’s one of those Thursday’s where no matter how well I prep, the big AI labs are hell bent to show up before each other. Alibaba dropped Qwen 3.6 with Apache 2, confirming their commitment to Open Source, then Anthropic released Claude Opus 4.7 (not quite Mythos) and OpenAI followed with a huge Codex update that includes Computer Use among other things. The highlight of Computer User is the background usage, more on that below. This is all just from today!Previously in the week we had 2 incredible 3D world generators, Lyra 2.0 from Nvidia and HYWorld 2 from Tencent, Windsurf dropping 2.0 version with Devin integration and Google releasing a Gemini TTS, with over 90+ languages support and incredible emotions range, and Baidu open sources Ernie Image, rivaling Nano Banana. Today on the show we had 3 awesome guests, Theodor from Cognition joined to cover the new Windsurf, Kwindla is back on the show to talk about “the side project that escaped containment” Gradient-Bang, a multi agent, voice based space game and Trevor from Marimo joined to talk about pairing your agents with a Marimo notebook. Let’s dive in! 👇 ThursdAI - We’re over 16K on YT today, my goal is to get to parity with Substack, please subscribe. Codex can now really use your computer: OpenAI updates Codex with CUA, Image Generation, Browser, SSH (X, Blog)Codex from OpenAI has been the major focus inside OpenAI for a while now. We’ve reported previously that OpenAI is closing down SORA and other “side-quests” to focus, and that they will join Codex, ChatGPT and the Atlas browser into one “superapp” and today, it seems, that we’ve gotten an early glimpse of what that app will be. The Codex team (which seems to be growing from day to day), have been on a TEAR feature wise lately, trying to beat Claude Code, and they pushed an update with a LOT of features and updates, among them a new memory system, internal browser and image generation. The highlight for me though, was absolutely the polished computer use experience. Computer use is not new, Claude has a computer use feature flag, many others. Hell, we told you about computer use with Open Interpreter, back in Sep of 2023. But, this.... this feels different. You see, OpenAI has quietly purchased a company called Software Apps Inc, that almost launched a macos AI companion a year ago called Sky. This team is obsessed with Mac, and somehow, they were able to build a magical experience, a huge part of which, is the fact that they are controlling the mac, in the background. This is like black magic stuff. You work on one document, Codex clicks buttons and does things in another, without interrupting you. You may ask, Alex, why do you even care so much about computer use, when most of the work happens in the browser anyway, and Claude (and Codex) can control my browser anyway? Well, true, but not ALL work is happening there, for example, file system integration. It’s notoriously big part of browser automation that fails, when you need to upload/download files. I’ve spent countless cycles trying to get this to work with OpenClaw, and this, just does it. This closes the loop between knowledge work in the browser (yes, this thing can use your browser) and the broader OS. It’s so so polished, I truly recommend you try it. It’s as easy as @ tagging any app that you have running and asking Codex to do stuff there. Pro Tip: Enable fast mode for a much smoother experience. Anthropic Opus 4.7 is here, not quite Mythos, 64.3% Swe-bench Pro, tuned for long running tasks (X, System Card)What is there to say? Is this the model we expected from Anthropic after releasing the news about Claude Mythos last week? no. But hey, we’ll take it. I new Claude Opus, with a significantly improved multimodality capabilities, and a long horizon coding task improvements? For the same price? Well, not quite! Apparently, this model could be a “from scratch” trained model, given that the tokenizer (the thing that converts words into tokens for the LLM to understand) is a different one. It also uses 1.3x more tokens for the same tasks, which means, that the new and default model from Anthropic became effectively more expens