Meituan's LongCat 2.0: Fully Trained and Inferenced on Chinese ASICs
bycloud · youtube · 2026-08-09
A video review covering Meituan's latest flagship model, LongCat 2.0. The release features significant architectural improvements, with a major twist: all training and inference were entirely executed on Chinese ASICs, potentially challenging NVIDIA's moat.
- Availability: Model weights are open-sourced on HuggingFace, accompanied by an official blog.
- Research: A related paper on LongCat Sparse Attention was also released.
More from Infra
- Together AI Benchmarks Kimi K3: Ranks #1 Across Multiple Inference Tests — togethercompute · 2026-08-09
- Satire: SD Cards at $2/GB in 2026 Highlights AI Storage Costs — felpix_ · 2026-08-09
- Home Assistant 2026.08 Adds Official llama.cpp Integration — ngxson · 2026-08-09
- DeepSeek Local Deployment: Troubleshooting Severe Speed Drop with Speculative Decoding — Easy_Werewolf7903 · 2026-08-09
- Fixing Black Video Outputs with MiniMax H3 on AMD GPUs — Present-Guitar-3967 · 2026-08-09
- Enabling PCIe P2P on Consumer Nvidia GPUs Boosts LLM Throughput by 25% — BidonPomoev · 2026-08-09