Kimi K3 vs DeepSeek V4 Pro vs GLM-5.2: trillion-scale open models compared

Moonshot AI's Kimi K3, DeepSeek V4 Pro, and Zhipu AI's GLM-5.2 now hold the top of the open-weight leaderboard; all three are sparse MoE models with million-token context windows aimed at long-horizon coding and agent workloads, compared on benchmarks, license, and serving cost.