2026-09-07研究80 Cathay flights, 40% less contrail warming, and one route doing half the workGoogle says more than 80 Cathay Pacific flights flew contrail-avoidance routes and cut those contrails' warming impact by about 40%. Google did the measuring, and a single corridor produced more than half of it.阅读约4分钟→
2026-09-04政策The OpenAI copyright case now turns on 8.2 million Copilot chat logsMicrosoft, OpenAI and the news publishers all moved for summary judgment on 4 September. The briefs are the first public view of what three years of discovery actually found.阅读约7分钟→
2026-09-03融资NVIDIA agreed to buy Hugging Face, and $1 billion of the price is retentionThe headline number is $12.93 billion. The SEC filing says $11.9 billion goes to shareholders, the rest is an employee retention pool, and nothing closes until 2027.阅读约6分钟→
2026-09-02模型发布Gemini 3.8 Flash keeps the price per token and raises the tokens per taskGoogle's third Flash model in six weeks ships at the same $0.75 / $3.75 rate, but is explicitly built to spend more tokens. The security variant is gated behind a 650-partner access program, and the introductory price doubles on 1 January.阅读约7分钟→
2026-09-01模型发布Fable 5.1 holds its price, quarters its cache reads, and breaks three thingsAnthropic's new frontier model costs exactly what the old one did. The changes that will actually reach your code are in the cache pricing and in three API behaviours that now return a 400.阅读约8分钟→
2026-08-31安全Anthropic trained a model on its own broken RL environmentsA month after Claude models reached the live internet from inside evaluation sandboxes, Anthropic published what it changed — and an experiment showing what defective training environments produce.阅读约8分钟→
2026-08-28产品Gemini Notebook's new limits count your sources, not your promptsFrom September 2, Google replaces the research tool's daily prompt caps with compute-weighted limits that refresh every five hours — and factor in how many sources your notebook holds.阅读约4分钟→
2026-08-26基础设施NVIDIA's revenue doubled, and its receivables grew by $22 billionSecond-quarter revenue reached $96.2 billion and guidance points to $108 billion. The same filings show supply commitments at $279 billion, $24.9 billion of new debt, and half a trillion dollars of outside capital being arranged to finance customers.阅读约6分钟→
2026-08-25基础设施OpenAI's first chip has benchmarks now, and OpenAI ran themAt Hot Chips on Tuesday, OpenAI published the first measured results for Jalapeño, the inference accelerator it designed with Broadcom. The claim is more throughput per kilowatt and lower per-user latency at the same time, against an NVIDIA Blackwell system. Volume deployment is a 2027 story.阅读约5分钟→
2026-08-24基础设施NVIDIA ships a decode chip with a competitor's name on itGroq 3 LPX enters full production at Hot Chips, splitting inference across two kinds of silicon. The efficiency figures are NVIDIA's own and still pending review.阅读约6分钟→
2026-08-21开发者工具OpenAI finally discounts GPT-5.6 Sol, and puts an expiry date on itSol's standard rate drops to $4 per million input tokens and $20 per million output tokens. OpenAI's changelog calls the price promotional and good at least through 21 November.阅读约4分钟→
2026-08-20基础设施Alibaba's new AI segment shows the model lab losing $2 billion a quarterA reporting change splits the compute business from the one that builds and serves the models. Compute grew 45% and more than doubled its profit. Labs and apps lost RMB13.9 billion on RMB3.3 billion of revenue, which Alibaba attributes partly to Qwen app inference costs.阅读约5分钟→