Monday, November 16, 2026
BP·InfoAI Briefing

The AI news that matters, explained for business and IT professionals.

Models & Research

Meituan releases open-weights LongCat-2.5 agent model, trained and served on Chinese hardware

Meituan has released the weights of LongCat-2.5, a 1.6-trillion-parameter mixture-of-experts model for coding and agent tasks, under an MIT license, The Decoder reports. By Meituan’s own tests it scores 88.4 on Terminal Bench v2.1, just behind Claude Opus 5. Its sample deployment runs on 194 Huawei Ascend chips and the guide does not mention Nvidia GPUs.

Why it matters

The release arrives as Washington debates tighter chip export rules, and it is presented as a model that did not need Nvidia hardware. Caveats apply: most results are Meituan’s own, the model trails Claude Opus 5 on the harder Terminal Bench v4.0 (38.5 versus 53.9), and The Decoder’s comparison leaves out newer Western models.

For business & IT

Teams evaluating open-weights models for coding agents can test LongCat-2.5 on their own tasks rather than trusting vendor benchmarks, and should review data-residency and supply-chain policies before using a Chinese-hosted API.

#meituan#longcat#open-weights#huawei-ascend#agents

More in Models & Research

See all →

Claude Fable 5.2 makes images by writing code that places simulated brushstrokes

According to The Decoder, Anthropic’s Claude Fable 5.2 creates pictures with Python programs that drive a simulated paint engine, rather than computing pixels like diffusion models. Anthropic employee Jake Eaton says one photorealistic parking-garage scene took roughly 700,000 strokes. The model uses no external image model or traced source images.

The Decoder