Meituan open‑sources 1.6 trillion‑parameter LongCat‑2.0 model, Moore Thread adapts it for MTT S5000 GPU

Meituan has released "LongCat‑2.0", a 1.6 trillion‑parameter mixture‑of‑experts (MoE) language model designed for agentic coding tasks. The model activates about 48 billion parameters per token and supports a native 1 million‑token context window. Benchmarks show it surpasses GPT‑5.5 on SWE‑bench Pro and performs competitively with Google Gemini 3.1 Pro, though it trails top‑ranked systems on broader agent benchmarks.

Moore Thread announced that its MTT S5000 AI‑compute GPU card and the MUSA software stack have been fully adapted to run LongCat‑2.0. The adaptation covers model loading, inference engine launch, key‑operator optimisation, deployment validation and accuracy checks, enabling stable, high‑efficiency inference on the hardware. The GPU’s native FP8 support and large memory bandwidth help handle the model’s long‑context and multi‑round interaction workloads, shortening the deployment cycle for enterprise AI coding, agent, and knowledge‑base applications.