ERNIE-5.0-Preview-1203 from @Baidu_Inc just scored 1451 on LMArena’s text benchmark, taking #1 among Chinese models. It’s a +23 jump from the previous release, and the gap between it and the next Chinese model (Qwen3-max-preview at #22) is significant. It stands out in creative writing and in handling complex, multi-step prompts, which has real implications if you’re building content pipelines or working with nuanced instruction flows. See the results for yourself 👇