METR finds Claude improving about 2x faster as Anthropic's IPO range slips to $1.1–1.5T

METR, an independent AI evaluation group, reports that the pace of Claude capability gains has roughly doubled compared with a year ago. In METR's analysis, a large share of that acceleration comes from Anthropic using AI systems in its own research and development. This kind of recursive dynamic has long been flagged by safety researchers as a key risk indicator.

The financial picture is similarly extreme. Anthropic posted $17 billion in third-quarter revenue alongside a $23.3 billion net loss, reflecting enormous compute spending. Prospective IPO buyers are now discussing a valuation as low as $1.1 trillion, with a range up to $1.5 trillion, well below earlier talk of $2 trillion. The listing is planned for mid-November.

The juxtaposition matters: investors are being asked to price a company whose growth and losses are both accelerating, and whose fastest gains may come from a feedback loop that is hard to oversee. OpenAI's private $1.25 trillion valuation this week gives the market a direct comparison point. Anthropic's product cadence continues regardless, including a Claude Code reference pattern on Bedrock AgentCore and an engineering post on memory 'dreaming.'

Skeptics note that commenters fear the speedup 'could outrun our ability to understand and control these systems.' Developer sentiment is also mixed: Goldman Sachs pulling Claude Code from bankers, the Apiiro secret-leak report, and Anthropic's 12-month credit expiry policy all drew criticism this week.