Meituan open-sources LongCat-2.5, a 1.6-trillion-parameter agent model trained on Chinese chips
🤖 AI-generated content — The title and summary were produced automatically by artificial intelligence, without human editorial review.
Meituan’s LongCat team released weights for LongCat-2.5, a mixture-of-experts model for long agent tasks and coding, trained entirely on Chinese hardware with no Nvidia GPUs mentioned. It scores 88.4 on Terminal Bench, just behind Claude Opus 5; a sample run uses 194 Huawei Ascend chips. About 48 billion of 1.6 trillion parameters activate per token, with API limits of one million input tokens and 128,000 output tokens, plus text, image, and video input.
Continue in the app — vote & join in ➔Source: The Decoder (AI) · via ahirlevel.hu