AI;DR: AI news, ranked every 30 minutes
03AI;DRTUESDAY · NOVEMBER 2026
- anthropicAnthropic’s amended filing puts Q3 revenue at $16.96B, while NYT reports $16.8B, but growth appears to have flattened. It swung to a $567M GAAP operating loss as compute costs rose 78% to $9.86B.+1
- anthropicGoldman Sachs is removing Claude Code and GitHub Copilot from traders and bankers, leaving access to engineers. The bank cites productivity and risk management after CIO Marco Argenti praised vibe coding in July.
- regulationChina’s MOFCOM and MOST proposed export controls covering 40 technologies, including 23 AI technologies, on October 16, 2026. Pre-export registration is already required, while controls on frontier model weights and agents remain open for comment until November 15.
- llmAstra can natively read and write neural-network weights to improve Qwen. The result suggests frontier LLMs handle weight editing far better than humans or conventional algorithms.
- fundingOracle plans to raise $20B through $10B in stock and $10B in unsecured bonds to fund AI data centers, including for OpenAI, and repay debt. Its shares fell more than 6%, leaving them about 54% below the year’s high.
- agentLongCat-2.5 is a 1.6-trillion-parameter model with native multimodality for agentic tasks. Training was completed entirely on domestic chips, highlighting China’s push for AI hardware independence.
- agentMomentic launched Mo, an AI QA engineer that tests apps by using them and has caught 117,010 bugs. It supplies reproduction steps and evidence while maintaining plain-English tests.
- llmDeep Principle released Hydrogen-1.0, a 753B open-weight scientific model post-trained from GLM-5.3. It scored 38.3% on FrontierScience-Research, targeting scientific discovery.
210 storiesUpdated 29m ago