Z.ai (Zhipu AI) released GLM-5.3, achieving a 50% improvement in coding performance over GLM-5.2 by reusing the same base model and focusing entirely on post-training. The model surpasses US rivals in CyberGym vulnerability discovery, nearing Anthropic's Mythos 5, though it trails in exploit execution. Open weights will be available in two weeks; until then, access is limited to a $18/month coding plan.
Alibaba officially released Qwen3.8-Max (2.4 trillion parameters) and Qwen3.8-27B, publishing the flagship model's open weights for the first time. Qwen3.8-27B fits frontier-level benchmarks onto a single consumer gaming GPU, and the Qwen family now tops Hugging Face's 2026 open model rankings, ahead of Meta and Google.
DeepSeek released V4 Pro (0813) with enhanced agentic capabilities for multi-step task execution, benchmarked against GPT, Claude, and Gemini rivals. The company also confirmed price increases for both V4-Flash and V4-Pro models, reflecting growing demand in China's competitive AI market.
Europe postponed implementation of the AI Act until 2027 to boost investment, infrastructure, and capabilities needed to compete with the US and China. The EU's AI Omnibus regulation (EU 2026/1744) entered into force on July 27, 2026, aiming to simplify existing AI provisions.
The White House is preparing to expand its AI policy framework, with sources indicating open models may soon be formally included in an updated framework. Separately, the White House has declined to publicly release its voluntary frontier AI model testing framework, drawing criticism over transparency from policy analysts.
Broadcom reported $30 billion in AI chip bookings while facing a $370 billion financing challenge, with Apollo and Blackstone supporting a first $35 billion tranche for over 1 GW of capacity. The AI chip-financing vehicle could carry $370 billion in senior debt by mid-2029, underscoring the massive capital demands of AI infrastructure buildout.
Vals AI closed a $40 million Series A at a $400 million valuation led by Andreessen Horowitz, after finding that frontier AI models fail 52% of real-world finance analyst tasks. The startup aims to solve the "Akerlof problem" in AI, where labs know their models' weaknesses far better than the enterprises buying them.