OpenAI Astra Solves 10 Math Problems for $2,000 | Grok 4.6 Launches with 1.5T Parameters | EU AI Act Takes Effect

🧮 OpenAI Astra Conquers Ten Mathematical Breakthroughs for Just $2,000

On August 1, OpenAI officially previewed its next-generation model family, codenamed Astra, now in closed beta. In internal testing, Astra achieved ten new mathematical breakthroughs spanning sphere packing (the first improvement since 1978), group theory, and quantum complexity. The findings are compiled into a 249-page paper with machine-verifiable Lean proof certificates.

Even more remarkable is the cost — total inference spend was approximately $2,000, averaging $200 per problem, roughly equivalent to a North American graduate student's weekend stipend. AI is no longer merely "solving problems" — it is beginning to produce genuine mathematical discoveries.

The mathematics community has responded with enthusiasm. As one scholar noted: if AI can resolve decades-old problems at such low cost, the paradigm of scientific research may be undergoing a fundamental shift. Astra isn't just another parameter checkpoint — it represents a qualitative leap in capability.

🚀 Musk's Grok 4.6 Launches August 7 — 1.5 Trillion Parameters Trained on Rocket Data

At SpaceX's Q2 2026 earnings call, Elon Musk announced that Grok 4.6 will officially launch around August 7, featuring 1.5 trillion parameters with significant advances in supervised fine-tuning (SFT) and reinforcement learning (RL). An even more ambitious successor — Grok 4.7 at 2.1 trillion parameters — is slated to follow within 3–4 weeks.

The standout differentiator: Grok 4.6 is the first AI model trained on over 20 years of SpaceX operational data, spanning the full rocket manufacturing, launch, and recovery pipeline. This proprietary data moat is one no other AI company can replicate. Musk's strategy is clear — exclusive data + massive scale + SpaceX engineering — charting a course in the AI arms race that no competitor can easily follow.

🤖 Ant Lingbo Raises 1.5 Billion Yuan — Embodied AI Frenzy Accelerates

Ant Group's embodied AI subsidiary, Ant Lingbo, has confirmed the launch of its first independent funding round, targeting 1.5 billion yuan, with a second round planned before the end of 2026. Rather than building robots, Lingbo focuses exclusively on the robot "brain," having already released over 10 LingBot series models pre-trained to adapt across 20+ configurations spanning 17 robot brands.

The broader sector is surging: Unitree Technology begins its STAR Market offline subscription on August 10; AGIBOT has initiated its Hong Kong IPO process; Poq Robotics recently closed a billion-dollar funding round. In the first half of 2026, embodied AI financing in China totaled approximately 43.8 billion yuan, with brain-focused companies attracting the largest share.

The timing of Ant Lingbo's independent fundraise is telling — Ant Group's AI spending has grown so vast that internal compute budgets have begun competing with one another. Spinning off for external capital not only brings in funding but signals that embodied AI is now viewed as a core track deserving its own independent valuation.

⚖️ EU AI Act Core Provisions Take Effect — "Watermark Red Line" Now in Force

On August 2, the core provisions of the EU AI Act — the world's first comprehensive AI law — officially took effect. While obligations for high-risk systems are deferred to December 2027, transparency requirements for general-purpose models, machine-readable labeling of AI-generated content (the "watermark red line"), and chatbot disclosure obligations are now enforceable.

Reactions from U.S. tech giants are sharply divided: Meta publicly refused to sign the EU's Code of Practice for General-Purpose AI, drawing opposition from 45 European companies; meanwhile, the White House finalized a voluntary safety assessment framework for frontier AI models, inviting Meta, Anthropic, OpenAI, and others. The Atlantic is charting two very different regulatory paths.

For Chinese AI companies, the EU Act means compliance is now a prerequisite for global expansion. AI-generated content labeling and model transparency disclosure will become hard requirements — companies that invest early in compliance capabilities will gain a meaningful competitive edge in the global arena.