TREE NEWS update: Meituan open-sourced LongCat-2.5 under the MIT license, allowing commercial inference deployment and further training without additional authorization. The native multimodal model for long-horizon agent tasks has 1.6 trillion total parameters, activates about 48 billion per token, and supports a one-million-token context. Meituan said it also optimized the architecture and deployment for domestic compute chips constrained by memory and bandwidth.
Meituan Open-Sources LongCat-2.5 With 1.6T Parameters
AI take
The MIT license is the real story here: it removes the legal friction that usually slows commercial adoption of large models, and Meituan is effectively buying developer mindshare by giving away a frontier-scale system. The explicit optimization for domestic compute chips constrained by memory and bandwidth matters more than the parameter count, since it signals the model is built for deployment environments where top-tier hardware isn't available. What's worth watching is whether the sparse activation design and long-context support translate into real agent workloads, and whether other Chinese labs follow with similarly permissive terms.
Generated by AI for reference only.
Share on WeChat
Open WeChat → Scan → then tap "…" to send to a chat or Moments.
Tap "…" in the top-right corner to send to a chat or share to Moments.
Related News
27m ago
SoftBank’s Total OpenAI Investment Reached $54.6B by End of Q2
42m ago
Golden Rooster Film Festival Debuts AI Unit With 72-Hour Filmmaking Challenge
52m ago
AI Short Drama Team Warns Beginners: High Barriers to Entry
1h ago
TSMC October revenue hits NT$586.89B, up 59.7% year on year
1h ago
Jiangsu Targets Over 10 GW New Energy Storage by 2030
2h ago
Anthropic Pulls IPO Filing Minutes After Release; $69.3B Round Details Leak