Meituan trains 1.6T coding model on Chinese ASICs, bypassing Nvidia

Meituan has released LongCat-2.0, a 1.6-trillion-parameter Mixture-of-Experts model designed for agentic coding tasks. Notably, the model was trained entirely on domestic Chinese AI ASICs, avoiding Nvidia hardware. LongCat-2.0 features a novel sparse attention mechanism for million-token contexts and demonstrates competitive performance against leading closed-source models on software engineering benchmarks. AI

IMPACT Demonstrates viability of training large models on non-Nvidia hardware, potentially diversifying AI infrastructure and lowering costs.

RANK_REASON New frontier model release from a major tech company (Meituan) with significant infrastructure and technical details. [lever_c_demoted from frontier_release: ic=1 ai=1.0]

Read on dev.to — LLM tag →

• Claude Opus 4.8
• DeepSeek Sparse Attention
• Gemini 3.1 Pro
• GPT-5.5
• Hermes
• Huawei Ascend
• Hugging Face
• LongCat-2.0
• Meituan
• Nvidia
• OpenRouter

• model release
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →