OpenAI's next model solved 10 open math problems this week while the EU's AI enforcement rules went live: August is already reshaping what AI can do and who's accountable for it.
OpenAI's next model Astra solved 10 open math problems, with verifiable proof
On August 1, OpenAI revealed that an internal version of its upcoming model, Astra, solved ten previously unsolved problems in mathematics and theoretical computer science. The proofs are machine-checkable in Lean, meaning they can be independently verified. This is not AI generating text that sounds like a proof. This is AI producing proofs that hold up under formal mathematical verification.
Why it matters: This is the clearest signal yet that AI has moved beyond being a productivity tool into territory that can contribute genuine new knowledge. If AI can crack open problems in pure mathematics, the implications for science, medicine, and engineering are enormous.
EU AI Act enforcement is now live, chatbots must say they're AI, deepfakes must be labeled
As of August 2, 2026, the European Commission began formally enforcing the EU AI Act's transparency requirements. Any chatbot or conversational AI must immediately disclose it is an AI at the start of every interaction. AI-generated images, audio, and video (including deepfakes) must carry visible labels and machine-readable marks so they can be detected. Fines for violations can reach €35 million or 7% of global annual revenue, whichever is higher.
Why it matters: This is regulation with real teeth, not guidelines. If you interact with any AI assistant (or create AI content) in Europe, the rules changed this week. Other governments are watching closely, and similar rules are expected to follow elsewhere.
Maharashtra just approved India's most ambitious state AI policy
Maharashtra approved its AI Policy 2026, targeting ₹10,000 crore in investment and 1.5 lakh new jobs by 2031. The plan includes training 2 lakh youth through a new Maharashtra Centre for Advanced AI Training (MCAT), establishing six AI Excellence Centres and five AI Innovation Cities, providing 2,000 GPUs for research use, and a ₹500 crore AI Startup Venture Fund split equally between the government and private sector. Around 5,000 MSMEs will also get direct financial support to adopt AI tools.
Why it matters: Maharashtra is home to Mumbai and Pune: India's tech and financial heartlands. This policy sets a benchmark other states will now have to compete with. If you are a student or entrepreneur in India, this creates real infrastructure and funding pipelines worth knowing about.
Frontier AI just got 60% cheaper: OpenAI cuts GPT-5.6 prices
OpenAI cut the price of GPT-5.6 Terra by 60% on July 30, making its mid-tier frontier model substantially more affordable for developers and businesses. This follows the broader pattern of AI infrastructure costs dropping faster than most forecasts predicted. The cheaper pricing applies via API, meaning apps built on GPT-5.6 Terra can now pass those savings directly on to users.
Why it matters: When the cost of calling a frontier AI model drops 60% in one move, the barrier to building AI-powered products falls with it. If you shelved an idea because of API costs, now is a good time to revisit it.
Google's Gemini 3.6 Flash takes the top spot in AI video generation
Google's new Gemini 3.6 Flash model currently leads the independent video generation leaderboard with an Elo score of 1527, a 45-point gap over the second-place model. This builds on Google's push to challenge OpenAI across every modality, not just text, but video, image, and multimodal tasks simultaneously.
Why it matters: Video generation is rapidly becoming the competitive frontier in AI. For creators, marketers, and educators, the quality ceiling for AI-generated video just went up again, and the tools are accessible to anyone.
AI attracted $202 billion in venture funding in 2025, nearly half of all global startup investment
New data confirms AI startups captured approximately $202 billion in venture capital in 2025, representing close to 50% of all global startup funding, up 75% from the year before. Anthropic closed a $65 billion round in May 2026 at a $965 billion valuation, making it the most valuable private company in the world. 79% of AI funding went to mega-rounds of $100 million or more, a sign the market has moved past experimentation.
Why it matters: The scale of this investment means AI infrastructure is being built whether the world is ready or not. The people who understand how to use these tools (not just those who build them) are the ones who will benefit most.
AI proving it can discover things, not just do things
OpenAI Astra's math breakthrough is the week's most consequential story, not because it fixed a bug or automated a task, but because it produced original knowledge that did not exist before. That is a different category of AI capability entirely. The next 12 months of Astra's development (and how other labs respond) will determine whether this is a one-off result or the start of a new era in which AI meaningfully contributes to research and science.
DeepSeek-V4-Flash-0731 drops + the model release pace keeps accelerating
DeepSeek released its V4-Flash-0731 model this week, adding to a market where major releases now come every few weeks. GPT-5.6 Sol also launched in limited preview on June 27. The cadence of frontier model releases has roughly quadrupled since 2023. For most people, this means the AI tools available to them are meaningfully better every quarter, even if the individual releases are hard to keep up with.