Saturday, May 23. Edge of AI, breaking one. Alibaba's Qwen team let a model run for 35 hours straight optimizing code for its own custom chip. That same model matches Claude Opus 4.6 and beats DeepSeek V4 Pro. And the IPO math for OpenAI and Anthropic just got a lot clearer. Let's get into it.
Alibaba, the Chinese e-commerce and cloud giant, released Qwen3.7-Max, a proprietary model built for long-running autonomous agent tasks [1]. The team let it run for 35 hours straight, autonomously optimizing code for the company's own custom silicon. The result: it matches Claude Opus 4.6 on benchmarks and beats Chinese rivals DeepSeek V4 Pro and Kimi K2.6. The same model also steered a four-legged robot in a demo. The signal: autonomous agent endurance is now a product differentiator, not just a research curiosity. Alibaba is betting that the lab that can sustain the longest autonomous run wins the enterprise coding workflow. My read: this is the first time a major lab has published a single uninterrupted agent session measured in days, not hours. That changes the expectation for what "agentic" means.
Different beat. On the IPO economics side, OpenAI and Anthropic are heading toward public markets, and the financial split between them is stark [2][unverified]. OpenAI is on course to accumulate hundreds of billions in losses before reaching positive cash flow around 2029 or 2030. Anthropic, by contrast, projects $10.9 billion in revenue for the second quarter of 2026, more than doubling Q1's $4.8 billion, and expects its first-ever operating profit of $559 million for that period. The difference traces to business model: Anthropic is enterprise-only, while OpenAI chases consumer scale with massive compute burn. The signal: the IPO market is about to price these two labs on opposite financial narratives. One is a growth story with a ten-year horizon. The other is profitable today. Which matters because the market will have to decide whether AI profitability comes from enterprise margins or consumer volume, and the answer isn't obvious.
Quietly. Google CEO Sundar Pichai now calls links and sources a "part" of search, not the foundation [4]. The wording is deliberate: Google is shifting from traffic distributor to AI publisher, keeping users inside its own ecosystem. The article argues that source selection is becoming a question of editorial power. The read: this is the kind of semantic shift that sounds like a PR tweak but actually rewrites the web's economic relationship with Google. If links are just a "part" of search, then Google's AI-generated answers don't owe the open web the same primacy it once had. That's a quiet constitutional change for the internet.
Three stories, three signals. The longest autonomous agent run on record, the clearest IPO financial split yet, and a Google word choice that redefines the web's place in search. Next week is when the market has to price the Anthropic-OpenAI divergence.
That is the edge for today.