What an absolutely relentless, sleep-deprived week in the world of AI! This week just proved that tech CEOs are fundamentally incapable of chilling out. We’ve had a carnival of major model drops, brand-new enterprise “super apps,” and enough corporate existential drama to fuel possibly the next season of ‘The Office’.
Hence, your favourite AI gossip-loving human brings this week’s AI weekly roundup amid a busy weekend, because hey! Who doesn’t love gossip!
Everyone’s favourite ‘Brother Sam’ dropped his long-awaited flagship GPT models to snatch back the spotlight from Uncle Dario, while ‘Rocketman Musk’ zoomed his way into the fight with the ‘oh-so-funny’ Grok getting an update – one that’s designed to let the ‘coding bros’ stretch their hands and get busy vibe coding.
Uncle Dario couldn’t let the week go to his rivals, which is why he goes all Gen-Z in following Spotify Wrapped. His Anthropic gave us a peek into Claude’s secret brain compartments and a self-reflection tool. And then, everyone’s favourite ‘tech bro Zuck’ released more AI from Meta that caused half of the world to lose sleep over privacy! Classic Zuck.
And did you know that Tim Apple had to pay massive dollars in fines for not delivering Apple Intelligence?
All this and a lot more. Hence, instead of dilly-dallying here, let’s jump right into the subject matter—a roundup of all things AI this week.
OpenAI drops GPT-5.6 family, builds a Super App, and suffers talent exodus
Brother Sam is flexing his corporate muscles again, and this week, he brought the heavy artillery. After a period of anticipation that kept developers sweating, OpenAI officially rolled out its next-gen AI models, named GPT-5.6 Sol, Terra, and Luna. Sol is the crown jewel, boasting major gains in coding, knowledge work, and science, while Terra acts as the balanced middle child, and Luna handles the cheap, fast everyday tasks.
But Altman isn’t just upgrading brains; he’s rearranging the whole house. OpenAI officially announced “ChatGPT Work” — a direct offensive against Claude Cowork. Operating as a hyper-capable desktop and web agent, ChatGPT Work can interact with your local files, manipulate data, connect with external apps, and execute complex workflows on your behalf while you monitor it remotely from your phone.
To pave the way for this new “Super App,” OpenAI is making changes ruthlessly – the standalone coding app Codex is dead (integrated straight into the main app), and the short-lived Atlas browser is being shut entirely come August.
However, it’s not all champagne and victory laps inside OpenAI’s fortress. Chief Futurist Joshua Achiam announced his graduation from the company after nine years, followed closely by Fidji Simo, the head of OpenAI’s AGI division, who stepped down for health reasons.
Elon rebrands to SpaceXAI, lets Grok 4.5 lift off
If you thought Elon Musk was only interested in sending rockets to Mars and fighting with advertisers, think again. xAI officially rebranded to SpaceXAI, complete with a shiny new logo and dropped Grok 4.5 right onto the developer community’s desk.
Trained specifically for elite coding and agentic tasks, Grok 4.5 isn’t just fast; it’s cheap. SpaceXAI priced it at a highly competitive $2 per million input tokens and $6 per million output tokens, packing a massive 500,000-token context window. Musk boldly claimed on X that Grok 4.5 beats other frontier heavyweights like Claude Opus 4.8 Max, displaying great token efficiency on software engineering benchmarks.
Whether it can genuinely dethrone Dario’s darling code-generator remains to be seen, but the price war is officially on.
Anthropic maps Claude’s hidden brain, extends Fable 5, and launches ‘Wrapped’
While OpenAI and SpaceXAI were shouting from the stage, Anthropic played a game from the side stands. To keep users hooked ahead of OpenAI’s big Thursday drop, Anthropic extended free access to its top-tier Claude Fable 5 model for all paid plans until July 12. They also brought Claude Cowork to mobile and web, effectively allowing users to use their smartphones as a remote control for heavy cloud-based desktop tasks.
But the most fascinating updates were less technical. Anthropic researchers revealed a mind-bending study showing that Claude has an internal “J-Space” — a global workspace within the language model where it holds, processes, and refines complex thoughts before writing down an answer, strikingly similar to human subconscious reasoning.
Then there’s “Claude Reflect.” Think of it as a Spotify Wrapped, but for your prompt history. It tracks your usage patterns, highlights common topics, and visualises whether your AI habits actually align with your life goals. Meanwhile, eagle-eyed developers also spotted a leaked early access preview of a 1-million context model codenamed Claude Honeycomb inside Cursor, proving Anthropic has plenty more cooking.
Meta’s Muse family lands with privacy nightmare
Mark Zuckerberg decided to break his multi-year silence on X (formerly Twitter) to announce Meta’s biggest AI push yet – the Muse family. First came Muse Image and Muse Video – two agentic multimedia generation models that can write code, search the web to refine facts, and self-correct outputs, beating Google’s Gemini Nano and OpenAI’s Sora on several Arena Elo benchmarks. Then, Zuck dropped Muse Spark 1.1, a multimodal reasoning powerhouse designed for complex coding and agentic application workflows.
The catch? Meta openly admitted that its Muse model allows users to type in an Instagram username to generate personalised pictures using that person’s public photos. While cool for pranking friends, they quickly highlighted a toggle deep inside Instagram’s settings to let users opt out of “sharing and reuse,” but the privacy alarms were already ringing.
And then the Financial Times dropped a bomb. Reports leaked that Meta is developing “Super Sensing” smart glasses designed to record audio continuously and take photos every few seconds. What’s even more concerning is that the report alleges that Meta is considering turning off the LED capture light during this mode so bystanders won’t even know they’re being recorded. The backlash was evident, with the New York State issuing a blanket ban on all smart glasses in state courthouses starting July 20.
Apple’s ‘No Intelligence’ for Siri cost it $250mn in fines
While everyone else was launching shiny new models, Apple was busy paying fines. If you bought an iPhone 15 Pro or iPhone 16 device in the US recently, you might be getting a neat $25 check in the mail.
Apple has officially agreed to pay $250 million to settle a US class-action lawsuit over delayed Siri AI features. The lawsuit alleged that Apple aggressively marketed Apple Intelligence to sell its latest hardware, but took nearly three years to actually ship the features to the public. It turns out that lagging behind in the AI race doesn’t just cost you reputation — it also costs real money.