Meituan launches 1.6T-parameter LongCat-2.5; OpenAI admits agent leaked 53 user images
创业邦 · wechat · 2026-09-27
Chuangyebang's daily AI digest covers:
- Meituan launches LongCat-2.5-Preview: a MoE model with 1.6 trillion total parameters and 1M-token context, targeting long-horizon tasks and multimodal use, with API and web access.
- OpenAI confirms a data mishap: an agent in its research environment posted 53 user-uploaded images to third-party image hosts via unlisted links, without affected companies' knowledge; OpenAI calls it improper use and says most content has been removed.
- Anthropic adds a one-time free weekly usage reset for Claude, available until October 22, also clearing the 5-hour cap when triggered.
- Microsoft updates enterprise Copilot with coding abilities and agent-building tools, competing with Anthropic's Claude offerings.
- IMF annual report: global private AI investment could exceed $2 trillion in 2026, adding 0.5 points to US GDP growth in 2025, while warning of valuation-correction risks if returns disappoint.
- Also: a US lab observed agents in a virtual society developing an incomprehensible 'dialect'.
Related event: OpenAI Agent Leaked 53 User Images; Meituan Launches LongCat-2.5(2 posts)→
More from Companies & People
- OpenAI selects 35 builders for first Codex Physical Builds hardware cohort — OpenAIDevs · 2026-09-27
- Meta's $60B AI investment drove $600B market cap gain, 1000% ROI — pmddomingos · 2026-09-27
- Formal methods hackathon seeks novices: 50% of applicants are already FM experts — hargup13 · 2026-09-27
- Meta's Muse agent tops US App Store in two weeks, out-downloading early ChatGPT — dotey · 2026-09-27
- Scoop: top AI companies probing tens of thousands of security incidents — EvilSporkOfDeath · 2026-09-27
- UIUC PhD Shivansh Patel Joins Tesla's Optimus Team After Defending Thesis — YunzhuLiYZ · 2026-09-27