1. Thinking Machines Lab Drops Inkling, a 975B Open-Weights Model
Mira Murati's Thinking Machines Lab released its first general-purpose model on July 15 — Inkling, a mixture-of-experts transformer with 975 billion total parameters and 41 billion active per token, supporting a 1M-token context window. It was pretrained on 45 trillion tokens of text, images, audio and video, and the weights are free to download on Hugging Face. Early benchmark chatter places it between Kimi 2.5 and 2.6 and ahead of Nemotron 3 Ultra — a serious open-weights entry from a lab that had shipped no model until now.
2. OpenAI's GPT-Red Beats Human Red-Teamers at Their Own Job
OpenAI introduced GPT-Red, an automated red-teaming model trained through self-play to hunt prompt-injection weaknesses. It succeeded in 84% of internal attack tests against 13% for human red-teamers, and OpenAI says it breaks nearly every model it is pointed at, up to and including GPT-5.5. Its findings were used to train GPT-5.6, cutting failures on one of the company's hardest prompt-injection benchmarks. OpenAI committed over 700,000 GPU hours to the effort, and says the red-teaming continues after launch rather than stopping at release.
3. Apple Sues OpenAI, Alleging Trade Secret Theft "at Every Level"
Apple filed suit against OpenAI on Friday, July 10 in federal court in Northern California, alleging OpenAI stole trade secrets to build its own hardware device. "OpenAI's nascent hardware business now rests on the shakiest of foundations, rotten to its core by its illegal reliance on misappropriated trade secrets," Apple wrote. Also named is io — the hardware startup founded by ex-Apple designer Jony Ive that OpenAI acquired last year for $6.5 billion — which Apple accuses of misleading an Apple partner into performing a confidential metal-finishing technique without authorisation.
4. China Clears Apple Intelligence — Powered by Alibaba's Qwen
China's Cyberspace Administration approved Apple Intelligence on Wednesday, July 15, ending a roughly two-year regulatory wait. Alibaba confirmed its Qwen model will power the generative features across Apple's operating systems for mainland users. The approval is a notable split in Apple's AI stack — Chinese iPhone owners get a domestically-built model rather than the one Apple ships elsewhere. Launch timing, feature parity and data handling all remain unresolved.
5. Anthropic and Blackstone Launch 'Ode' — a $1.5B Bet on Implementation
The joint venture Anthropic announced back in May now has a name and is live: Ode with Anthropic, a $1.5 billion AI implementation company backed by Blackstone, Hellman & Friedman, Goldman Sachs and others. It's built on Fractional AI, a consulting startup acquired in May. The thesis is that the next trillion-dollar AI business is deployment rather than models — closing the gap between an AI demo and a production system full of evals, edge cases and legacy integrations.
6. Hundreds of Economists: "We Must Act Now" on AI Job Displacement
In an open letter released Monday, July 13, over 200 economists warned that institutions must move now to address how AI could transform the economy and put many people out of work. The letter runs just four sentences, calling on leaders to "build the incentives, guardrails, and institutions needed to steer AI in a direction that complements humans and benefits society." It is deliberately short on prescriptions and long on urgency — the signatories are not predicting a specific outcome, they're arguing that the preparation window is closing.
7. Musk Quietly Bought a $1B Gas Turbine Company to Feed xAI
It emerged this week that SpaceX acquired APR Energy, a Jacksonville-based operator of trailer-mounted mobile gas and diesel turbines, for roughly $1 billion. The fleet exceeds 1 GW of capacity and can be deployed in months rather than the years a conventional plant takes — power for xAI's Colossus supercomputers training Grok. It's a pointed turn for a founder who built his brand on solar: the constraint on frontier AI is now electricity, and Musk just bought fossil generation to solve it.
8. WAIC 2026 Opens in Shanghai at Record Scale
The World Artificial Intelligence Conference runs July 17–20 in Shanghai, spanning a record 100,000 square metres across three venues in Pudong and Xuhui, with over 1,100 confirmed exhibitors and 112 forums. China's National Development and Reform Commission has said two major policy outcomes will be released during the conference. Hong Kong-listed AI model stocks including MiniMax and Zhipu rallied ahead of the opening.
// KEY TAKEAWAYS
Three fronts moved at once this week: open weights (Inkling's 975B MoE landing free on Hugging Face), safety tooling (GPT-Red automating what human red-teamers did worse), and raw physical capacity (Musk buying a gigawatt of trailer-mounted turbines because electricity, not algorithms, is the binding constraint). Meanwhile the money is shifting from models to deployment — Anthropic's $1.5B Ode bets the next trillion-dollar business is implementation, not another checkpoint. And the friction is turning legal and political rather than technical: Apple suing OpenAI over hardware secrets, China clearing Apple Intelligence only through a domestic model, and 200+ economists warning that the labour-market bill comes due before the institutions to handle it exist.