AI Daily · Oct 26, 2026 · Monday
BAD INFLUENCE
OpenAI and Anthropic cut LLM prices up to 80% within months
OpenAI cut GPT-5.6 Luna's price by 80% to $0.20 / $1.20 / $0.02 per million tokens on July 30, 2026, and Anthropic made its temporary Sonnet 5 launch pricing of $2 / $10 / $0.20 permanent on August 10. The author says the two companies traded price cuts in response to each other, with cache read costs making up 96% to 99% of agentic coding token spend. Models are increasingly interchangeable, and the author argues competitor pressure from open-weight models like DeepSeek V4 Flash explains the drops.
Hacker News · AI (150+ points)BlogRead the original Industry news
AI EraNewsOpenAI reportedly solves Siegel zero problem, cutting proof to five pages
AI Era reports that OpenAI has resolved the Siegel zero problem, a number theory question Zhang Yitang worked on for decades. According to the report, a human-and-GPT-6 effort over two weeks condensed the proof to five pages. The excerpt does not give further details of the proof or its verification.
Rohan Paul@rohanpaul_aiX2 more sourcesElvis Presley rights holders sue Anthropic over Claude training on 274 songs
Authentic Brands Group and several publishers filed a federal suit against Anthropic seeking about $41 million, alleging Claude was trained on 274 Elvis Presley songs. The complaint multiplies 274 works by the $150,000 maximum statutory damages per work for willful infringement and also targets Claude's outputs, citing a Claude response describing song rewrites that reused original structure and hooks. No court has tested these claims, and a 2025 ruling found training on lawfully purchased books to be fair use.
Papers & research
Greg Brockman@gdbOfficialGreg Brockman says OpenAI's fluid dynamics paper is technically unbelievable but understandable
Greg Brockman quotes a mathematician's thread arguing that OpenAI's blow-up solution paper combines convex integration with a physical fluid instability in a way with no precedent. Brockman says the paper is beyond what a human could have produced but not beyond human understanding, and that several groups of PDE experts say they more or less understand it.
ModelScope@ModelScope2022OfficialQwen Team releases BabelArena, a multilingual agent benchmark across 23 languages
Qwen Team introduces BabelArena, an open-source multilingual agent benchmark covering 23 languages, with the dataset and paper linked in the post. It reports that no single model leads every benchmark. Low-resource languages show more tool-use and control-flow errors, agents use up to about 2x more tokens in them without extra interaction rounds, and language consistency drops in structured-output tasks, with switches mostly to English.
