0
Applied AI·June 30, 2026·1 min read

Meituan open sources LongCat-2.0, the 1.6T, near-frontier agentic coding model that's been leading OpenRouter — trained entirely on Chinese chips

Share

A 1.6T-parameter, near-frontier coding agent trained entirely on Chinese chips and now open sourced compresses two trends—sovereign-scale training and open agentic tooling—into one move. For global devtool and infra teams, the assumption that top-tier coding models are Western and closed just broke.

Applied AI

Exclusive: Meta and OpenAI alumni seek $400m for new AI lab

Another $400M lab from ex-Meta/OpenAI talent means the frontier stack is still fragmenting rather than consolidating—talent and capital are betting there’s room for differentiated research agendas, not just scaling incumbents. If you’re an applied team, assume the model landscape in 18–24 months will be more crowded and specialized, not fewer-but-bigger, and architect for swap-ability rather than single-vendor dependence.

Applied AI

Meta is telling engineers to handle Claude Code and Codex with care

Tightening internal rules on Claude Code and Codex shows how seriously large platforms now treat inadvertent model distillation and IP leakage—coding assistants are no longer “just tools,” they’re potential data exfil paths. Any org building proprietary models should adopt similar policies this quarter: define which external AIs are allowed, where, and with what redlines around sensitive code and datasets.