ChatGPT Work launches as an AI that runs your entire workday, China's GLM-5.2 tops open-model rankings on Huawei chips, and the EU AI Act's enforcement deadline is just 23 days away.
ChatGPT Work is live: OpenAI ships an AI agent that does the full job, not just the answer
OpenAI launched ChatGPT Work on July 9, an agent built on GPT-5.6 that accepts a goal and autonomously produces finished spreadsheets, slides, documents, and code over hours without hand-holding. The Codex coding assistant merges into the same desktop app. Available now for Pro and Enterprise plans, rolling out to Plus and Business users this week.
Why it matters: This crosses a meaningful threshold, from AI as a tool you prompt to AI as a co-worker you assign. If it performs as shown, the way knowledge workers think about getting things done will shift. Worth testing yourself this week rather than waiting to read about it.
xAI releases Grok 4.5: three frontier labs shipped model updates in 48 hours
Elon Musk's xAI released Grok 4.5 with stronger performance on coding, multi-step agentic tasks, and long-horizon knowledge work. The release lands within 48 hours of Anthropic's Claude Sonnet 5 and OpenAI's GPT-5.6 family, making this one of the most compressed model-release windows the industry has seen.
Why it matters: Three of the world's top AI labs all improved their frontier models within two days. The competition is now moving so fast that last week's best model may already be outdated. If you're on a paid AI subscription, it's worth checking what's new in your plan right now.
China's GLM-5.2 is now the world's #1 open AI model, trained entirely on Huawei chips
Z.ai's GLM-5.2, a Chinese open-source model with 753 billion parameters and a 1-million-token context window, has topped the Artificial Analysis open-weight AI rankings, outperforming models from NVIDIA and DeepSeek. The model was trained on 100,000 Huawei Ascend chips with zero NVIDIA hardware, a direct response to US export restrictions. It is freely available on Hugging Face under an MIT licence.
Why it matters: The US restricted chip exports to China partly to slow its AI progress. GLM-5.2 is evidence that the strategy may have accelerated China's development of alternative infrastructure instead. The world's best free AI model now runs on Chinese silicon, and anyone can download it today.
Meta Muse and Google Video Remix bring creative AI into apps you already use
Two major consumer AI tools landed this week. Meta launched Muse, an image generation model inside Meta AI that can restore old photos, restyle rooms, and create surreal edits. Google Photos launched Video Remix, powered by Gemini Omni, offering cinematic relighting, background swapping, and artistic effects for personal videos. Both are rolling out to users now.
Why it matters: Creative AI is arriving inside apps that billions of people already open every day, not just in specialised tools for early adopters. The barrier to using AI to make something beautiful is now close to zero. If you use Google Photos or Meta AI, these features are available to explore today.
India's AI sovereignty challenge: caught between the US, China, and its own ambitions
A Bloomberg analysis published this week examines India's difficult position in the global AI race: dependent on US and Chinese infrastructure yet trying to build sovereign capabilities. The US-forced Anthropic Fable 5 ban raised pointed questions about technology dependence. Separately, India's government is actively considering a dedicated AI regulatory framework, moving beyond its current light-touch approach to address deepfakes, job displacement, and AI-generated harms.
Why it matters: India has the talent, the ambition, and a 1.4 billion-person market. But real constraints on compute and international access shape what Indian developers and businesses can actually build. How India navigates this moment sets the template for every emerging economy watching from the sidelines.
AI is getting dramatically cheaper, and that is when behaviour actually changes
Industry analysts are marking July 2026 as an inflection point: inference costs for capable AI models are falling sharply. Claude Sonnet 5 launched at $2 per million input tokens, down from rates that were prohibitive a year ago. The broader shift is from "bigger models" to "faster, cheaper, and reliable" AI, a transition that brings the technology within reach for schools, clinics, small businesses, and users in regions where cost was a barrier.
Why it matters: Price reductions are not just a business story. When capable AI costs less than a cup of coffee to run for an hour, access expands fundamentally. This is the economic shift that drives AI from a tool for the well-resourced to infrastructure for everyone, including students and professionals in India and across the Global South.
The EU AI Act enforcement clock is down to 23 days
The EU AI Act's transparency obligations take effect on August 2, 2026: 23 days from today. Companies operating in Europe must be able to explain their AI systems, disclose AI-generated content, and comply with rules on high-risk applications. Many businesses are still unprepared. India's own AI regulatory framework debate is heating up at exactly the same moment, watching Europe's enforcement model closely as a possible template.
NVIDIA + SK Hynix partner on next-gen AI memory • Fable 5 extended to July 12
NVIDIA and SK Hynix announced a multiyear partnership to co-develop next-generation memory aligned with AI infrastructure needs, the chip-layer story beneath every model you use. Separately, Anthropic extended Claude Fable 5 access for all paid users through July 12, giving a final window before the model fully transitions to Sonnet 5 and newer versions.