Meituan Orders Teams Off ByteDance's AI, Onto Its Own Models

Meituan this week sent a notice to all business teams restricting the use of ByteDance's Volcano Engine (Doubao) large language model in product and tech development. The directive requires departments to audit their current usage and plan migrations to Meituan's in-house model LongCat or the open-source model DeepSeek. Teams that face technical blockers and cannot move must submit a detailed justification and go through a separate approval process.

Meituan just released and open-sourced LongCat-2.0, a Mixture-of-Experts (MoE) model with 1.6 trillion parameters. Previously, some Meituan teams had adopted ByteDance's Doubao model because of its competitive pricing. By switching to its own LongCat-2.0, Meituan aims to reduce external AI dependence and promote its own model — but the move also comes as competition in local services heats up, with the company keen to keep its data and tech stack out of a major rival's hands.