AI news, 23 September: Anthropic and OpenAI launch cheaper high-performance models

Anthropic and OpenAI cut costs on new frontier models. On Tuesday, Anthropic released Claude Opus 5.5, which it says matches the performance of its top-tier Fable 5.1 on most tasks while costing about 40% less to run than the prior Opus 5, according to the company and Reuters. The model underwent external safety testing and showed strong results on internal alignment checks. Hours later, OpenAI expanded its GPT-6 lineup with Sol and Luna, priced at half the promotional rates of their GPT-5.6 predecessors and carrying many advances from the flagship Astra model in reasoning, coding and reliability. The dual launches highlight intense competition to make advanced AI more affordable even as industry leaders debate slowing capability releases for safety reasons.

Alibaba unveils China's most powerful AI chip and bigger model plans. Chinese tech giant Alibaba announced the Zhenwu V900 chip from its T-Head unit, claiming it delivers three times the performance of its predecessor and can scale in huge clusters. The company also outlined plans for next-generation Qwen models with 5 to 10 trillion parameters and a target of over 20 gigawatts of global data-centre capacity by 2032. Shares rose sharply on the news, Reuters reported, underscoring China's push for domestic AI hardware amid export restrictions and ahead of high-level US-China talks.

Snorkel AI hits $3.5 billion valuation on training-data boom. Data specialist Snorkel AI raised $350 million in a Series E round led by Insight Partners and S32, nearly tripling its prior valuation, according to Reuters and the company. Annualised revenue run-rate has surged past $350 million as frontier labs demand complex datasets and simulated environments. The funding will expand hiring and support for enterprise and government customers.

OpenAI claims AI solved 100+ maths problems and forms advisory panel. OpenAI said an internal model has resolved more than 100 long-standing open mathematical problems and formed an independent Advisory Group on Mathematics and Artificial Intelligence, hosted at the Institute for Advanced Study. The unpaid panel of leading mathematicians will advise on reviewing and communicating results but has no decision-making power over the company's research. The claims remain largely unverified by outside experts.

Most Americans say AI firms fall short on safety. A new Reuters/Ipsos poll found 73% of Americans worry that AI companies have not done enough to prevent serious harm to society. Thirty-nine percent view AI's societal impact as negative—the highest share recorded so far—and a majority favour slowing development, reflecting growing public caution amid rapid advances.