Meituan has fully open-sourced LongCat-2.0 and the inference code for domestic accelerator cards. LongCat-2.0 is the first trillion-parameter model to complete inference on a 50,000-card domestic compute cluster. Meituan says it co-optimized model architecture, chip adaptation and deployment strategies to run the trillion-parameter model stably and quickly on memory- and bandwidth-constrained domestic chips and older cards. The release includes BF16, FP8 and INT8 multi-precision builds and open-

Meituan has fully open-sourced LongCat-2.0 and the inference code for domestic accelerator cards. LongCat-2.0 is the first trillion-parameter model to complete inference on a 50,000-card domestic compute cluster. Meituan says it co-optimized model architecture, chip adaptation and deployment strategies to run the trillion-parameter model stably and quickly on memory- and bandwidth-constrained domestic chips and older cards. The release includes BF16, FP8 and INT8 multi-precision builds and open-sourced inference optimizations targeted at domestic compute platforms to enable deployment on existing hardware.