Monday, Nov 2, 2026
AI;DR
2026-11-02 · 16 stories- anthropicAnthropic’s amended IPO filing reports Q3 revenue of $16.96B, up 1,127% year over year, but a $567M GAAP operating loss. Compute and infrastructure costs rose 78% as commitments reached $895.4B.
- agentLongCat-2.5 is a 1.6T-parameter native multimodal model built for agentic tasks. Its training was completed entirely on domestic chips, highlighting China’s push for self-reliant AI infrastructure.
- regulationChina’s MOFCOM and MOST proposed export controls covering 40 technologies, including frontier model weights and agent architectures, while requiring pre-export registration immediately. The draft remains open for comment until November 15, 2026.
- openaiOpenAI will use its internal model that solved the Navier-Stokes equations as the foundation for its next GPT release. The solution came from 10,000 agents collaborating for about four days, and the model is expected soon.
- regulationThe BAAJA’s ratepayer protections could make data centers pay more than $50B for transmission over 10 years. Their bill could rise by another $100B or more if states also require funding for generation.
- codingA Proton memo tells engineers to prioritize AI speed and predicts most human code review will eventually end, although productivity has not risen yet. Proton says security reviews have increased despite AI-generated code.
- llmDeep Principle released Hydrogen-1.0, an open-weight model for scientific discovery, scoring 38.3% on FrontierScience-Research versus GPT-6-Astra’s 33.3% and 93.0% on GPQA-Diamonds. It is meant to broaden access to research AI.
- anthropicSources told The New York Times that Anthropic expects more than 3.5GW of compute capacity in 2026 and is expanding direct hardware procurement to Nvidia and AMD. The move supports its rapid AI infrastructure buildout.
- nvidiaMarvell agreed to buy optical-components maker Lumentum for $115B as it seeks to challenge Broadcom. Nvidia invested $2B in both companies, tying the deal to the broader AI networking market.
- fundingSouth Korea will nearly double government AI spending to 21.3T won starting next year. President Lee said the investment is intended to strengthen economic growth and fund AI and other new growth drivers.
- openaiOpenAI is merging ChatGPT’s chat and Work tabs, with ChatGPT and Codex drawing usage from the same allocation pool. The change simplifies access to both experiences under one usage system.
- safetyWIRED found about 400 ordinary-looking AI apps and websites linked to Chinese-language developers that secretly enabled violent, sexually abusive deepfake images and video. The discovery exposes abuse risks hidden behind benign interfaces.
- regulationThe White House told The Information it need not share the full draft of its voluntary AI testing framework. The Trump administration cited national-security and commercial exemptions for withholding key documents.
- llmZ.ai’s GLM-5.3-FlashX-Legal is now on OpenRouter, accepting text, images, and video while producing text. The legal-domain model has a 1,048,576-token context window and costs $0.37 input and $1.25 output per million tokens.
- An alleged Anthropic Mythos-class checkpoint leak on Hugging Face claims to offer a 6.8T-parameter machine learning model. The listing invites users to download, explore files, and run it, but the leak remains unverified.
- openaiSemiAnalysis launched GDPx, an agentic benchmark spanning 11 domains, including chip design, cybersecurity, RSI, bio, finance, legal, quant trading, software engineering, materials science, and alignment. Opus 5.5 led with 12%.