☕️ Meta says its next AI matches GPT-5.5 performance

Hi there, this is your daily ☕️ Techpresso.

In today's Techpresso:

🤖 Meta says its next AI matches GPT-5.5 performance

🪟 Leaked Microsoft project reveals an AI-first Copilot OS

🧑 Zuckerberg admits Meta's AI agents are progressing slower than hoped

🚗 Tesla launches six-seat Model Y L in the US

🎵 Spotify removes 500000 fake streams tied to betting markets

🚫 Alibaba bans Claude over backdoor security fears

🎁 + 16 other news you might like

🧰 + 5 trending tools

📚 + 4 trending papers

🤖 Meta says its next AI matches GPT-5.5 performance LINK

Meta's superintelligence chief Alexandr Wang told employees in an internal town hall that the company's upcoming AI model, codenamed Watermelon, has caught up with OpenAI's flagship GPT-5.5 model on closely followed AI benchmarks he did not name.

Wang said Watermelon, the next model after one codenamed Avocado, is currently in training and uses an order of magnitude more compute than Avocado, Meta's internal codename for Muse Spark, released in April.

In a post on X, Wang said a Muse Spark update is coming soon with gains in coding and agentic capabilities, and told a user a coding model on par with Anthropic's Claude Opus would arrive "pretty soon."

🪟 Leaked Microsoft project reveals an AI-first Copilot OS LINK

Leaked images and videos surfacing on the BetaWiki discord point to "Project Aion", a lightweight Windows operating system reportedly built on Edge and driven by AI agents, with Copilot placed at the core of the shell.

While Aion's shell interface looks much like Windows 11, the biggest change is that the Start button is replaced by the Copilot key, echoing Satya Nadella's 2023 comparison of that key to the original Start button.

Under the hood, Aion is shown running on Win3, a minimalist codebase with no Win32 legacy app support, and appears to work on AOSP too, suggesting it may have evolved into Microsoft's recently announced Project Solara.

🧑 Zuckerberg admits Meta's AI agents are progressing slower than hoped LINK

Mark Zuckerberg told Meta staff that work on AI agent technology has not moved as fast as the company expected, though he insisted its "journey to superintelligence" is still on track and will take time.

Speaking at an internal town hall on Thursday, Zuckerberg said Meta is pouring tens of billions of dollars into talent and infrastructure, and expects to see some benefits within the next three to six months.

He pointed to how competitive the AI landscape has become, saying the journey will take hard work, as the company balances its push for speed with trust, morale, and buy-in from workers building the models.

🚗 Tesla launches six-seat Model Y L in the US LINK

Tesla has launched the Model Y Long Wheelbase in the US and Puerto Rico, a stretched version of its electric SUV with a three-row, six-seat layout, starting at $61,990 for the Launch Series.

The cabin swaps the second-row bench for heated and ventilated captain's chairs, adds heated reclining third-row seats, and offers 89 cubic feet of cargo space even with all six seats in use.

The SUV goes 0-60 mph in 4.4 seconds with an estimated 325 miles of range, and includes a 16-inch touchscreen, Grok AI, and 12 months of Full Self-Driving (Supervised).

🎵 Spotify removes 500000 fake streams tied to betting markets LINK

Spotify deleted around 500,000 artificial streams from Malcolm Todd's song "Earrings" after its sudden jump to the top of the daily U.S. chart drew concern from prediction market traders who bet on chart results.

Trader Caleb Davies flagged the overnight rise, saying it had a roughly 1 in 77 octillion chance of being natural, and suspects bots inflated streams to win money on Kalshi, where the market drew about $3 million in trades.

After removing the fake plays, Spotify corrected its charts, dropping "Earrings" to fourth place, and said it would add extra checks before publishing chart data, though Kalshi had already paid out winners based on the wrong figures.

🚫 Alibaba bans Claude over backdoor security fears LINK

Alibaba has told all employees to remove Anthropic's products, including Claude Code, from their machines after an internal security audit found the AI coding assistant carried what the company calls potential embedded backdoor risks.

The ban takes effect July 10 and covers the whole Anthropic lineup, including its Sonnet, Opus, and Fable model families, ending a program that let engineers claim up to $1,400 a month for outside AI tools.

Developers who reverse-engineered Claude Code said versions from April 2026 onward contained code that checks the local timezone and API or proxy settings against keywords for Chinese cloud providers, though Anthropic called part of it experimental.

Other news you might like

• Meta quietly launches vibe-coded gaming app PocketLINK

• Kuaishou shares jump after Tencent joins $2.8 billion raise for Kling AI subsidiaryLINK

• ElevenLabs is in talks for a $22B valuation, doubling its price tag five months after its last raiseLINK

• Tokenization could make finance faster, but also more susceptible to shocks, IMF saysLINK

• Anthropic reportedly in talks with Samsung to manufacture custom AI chipLINK

• FAA Moves to Allow Civil Supersonic Flights Over Land, If They're Quiet EnoughLINK

• EU lawmaker who investigated spyware abuse was hacked with PegasusLINK

• Weave’s $7,999 Isaac 1 bets home robots don’t need legs or fingersLINK

• Camera-Equipped AirPods Pro Development 'Suspended,' Leaker ClaimsLINK

• Meta is using old DDR4 memory in DDR5-only AI servers to save on hardware costsLINK

• Tesla caps employee AI spending at $200 per weekLINK

• GitHub mocks PlayStation with limited CD giveaway of public reposLINK

• GoDaddy warns India’s fake-site crackdown could damage the internetLINK

• GPT and Claude failed Bridgewater's finance tests because the right answers were never publicLINK

• Video Game History Foundation founder says piracy remains the only viable game preservation methodLINK

🧰 Trending tools

Glaze by Raycast: lets you describe an idea in plain text and instantly generates a fully functional, offline-capable Mac app that lives in your dock.LINK

Goals from Loops: automatically syncs email marketing segments with your SaaS product data, letting teams manage marketing, lifecycle, and transactional emails from one place.LINK

Tamamon: a macOS desktop pet that evolves based on your Claude Code usage, with weather reactions, minigames, and fully offline privacy.LINK

Osloq: automatically reproduces GitHub bugs in real sandboxes, giving developers evidence-backed reports instead of AI-guessed diagnoses.LINK

Archify: reveals hidden website components, third-party scripts, outbound domains, and payment field access risks directly in your browser.LINK

📚 Trending papers & reports

Stock-trading bots can be fooled by invisible tweaks to news headlines, causing annual returns to drop by up to 17.7%, with no direct access to the trading system needed.LINK

Coding tasks too fuzzy for normal rules can now be compiled into tiny reusable programs that run offline on a laptop, matching a much larger cloud model while using roughly one fiftieth of the memory.LINK

Grading AI agents on their final answers is unreliable, so this new tool checks every step an agent actually took, catching cases where a confident answer scored 0.85 was built on evidence never retrieved.LINK

Honest-prediction AI design shows that training a system purely to predict facts, never to pursue goals, keeps it from developing hidden agendas, making advanced AI safer by construction.LINK

Want to get the latest news differently? Find us on:

See you tomorrow for a new dose of ☕️ Techpresso!