AI News Roundup: A 12% Expert Test, Anthropic's Numbers, and a $115B Chip Deal

A new expert-level benchmark tops out at 12%, Anthropic files its third-quarter results, Marvell agrees to buy Lumentum, and The New York Times sues Brave over scraped articles.

Here is the AI news from the last day or so that matters for small businesses using AI. Every item links to its source.

The hardest test yet. SemiAnalysis released GDPx, a benchmark of 206 long, expert-level tasks across 11 fields like chip design, finance and law. The best model, Claude Opus 5.5, scored 12%, and every model scored zero on finance and legal under strict grading. AI is a strong helper, not a senior expert working alone. (Littlewolf Labs News, AI Lately)

Anthropic shows its books. Ahead of its IPO, Anthropic's amended filing reports $16.96 billion in third-quarter revenue and a $567 million operating loss. (The Diary of AI)

OpenAI's round comes in light. OpenAI has raised about $11 billion of a round it first aimed at $30 billion, and now targets $20 billion. Separately, Aramco paid about $2.6 billion for 49% of Saudi Arabia's HUMAIN. (AI Lately)

A $115 billion bet on light. Marvell agreed to buy laser maker Lumentum in an all-stock deal, betting AI data centers move from copper to optics. (AI.info)

Who pays for words. The New York Times sued Brave, saying its search API resold scraped articles to AI companies. (The Diary of AI)

What it means for you: AI does best on focused, repeatable work with a person in the loop. Answering every call, qualifying the lead, and passing hot ones to a closer is that kind of job. That is how Ozkra works. Want to try it on your calls? Talk to us.