The mystery behind “Owl Alpha,” the anonymous stealth AI model that has dominated global developer leaderboards on OpenRouter for the past two months, has finally been solved. On Monday, Chinese delivery and local services giant Meituan officially unmasked itself as the creator, unveiling the system as LongCat-2.0.
The release marks a watershed moment in the global artificial intelligence race. LongCat-2.0 is a massive 1.6-trillion-parameter Mixture-of-Experts (MoE) architecture featuring a native 1-million-token context window. More importantly, according to VentureBeat, the model was trained entirely on a cluster of over 50,000 domestic Chinese Application-Specific Integrated Circuits (ASICs), proving definitively that near-frontier AI models can be scaled without relying on Nvidia GPUs.
Meituan has released the model architecture under a highly permissive, enterprise-grade MIT license on GitHub and Hugging Face, though the company noted that the full model weights are “coming soon.”
A Breakthrough in Autonomous Coding
While LongCat-2.0 generally trails premium frontier systems like Anthropic’s Claude Opus 4.8 on broad, general-knowledge benchmarks, it was explicitly designed to excel at autonomous software engineering.
The model utilizes a unique MOPD gate-routing mechanism that segregates tasks into specialized expert clusters: Agent Experts, Reasoning Experts, and Interaction Experts. This allows the system to coordinate deep logic, stable tool execution, and safe user interaction simultaneously. As a result, LongCat-2.0 narrowly exceeded OpenAI’s proprietary GPT-5.5 on the rigorous SWE-bench Pro software engineering benchmark, scoring 59.5 against OpenAI’s 58.6.
This hyper-focus on coding makes LongCat-2.0 a highly disruptive tool for enterprise development. With its 1-million-token context window, engineering teams can pass entire legacy codebases and modern SDK documentation directly into the model. The system can map dependencies, execute repository-level structural updates, and catch compilation bugs autonomously within local sandbox environments before generating a final pull request.
Aggressive Pricing and Zero-Cost Caching
To drive commercial adoption, Meituan has introduced an aggressive, innovative pricing structure. The standard “pay-as-you-go” API is priced at $0.75 per million input tokens and $2.95 per million output tokens. However, a limited-time promotional discount slashes these rates to $0.30 and $1.20, respectively, placing it among the cheapest top-performing models globally.
The true economic disruption, however, lies in Meituan’s caching architecture. Under the LongCat-2.0 framework, all context-cache hits are processed at no cost. In massive agentic environments where a coding assistant must repeatedly read and reference the same multi-million-token code repository over an extended session, standard architectures penalize developers by charging for repeated input context. Meituan’s zero-cost caching completely alters the operational economics of large-scale agent software development.
To manage the massive compute load this will generate on its ASIC clusters, Meituan is selling high-volume “Token Packs” via limited flash sales four times daily (at 10:00, 16:00, 21:00, and 23:00 Beijing Time). These packs stack on top of standard API accounts to accommodate the unpredictable compute bursts characteristic of autonomous development agents.
From Super App to AI Powerhouse
The release of LongCat-2.0 represents a massive strategic pivot for Meituan. Founded in 2010 by serial entrepreneur Wang Xing, the company evolved from a Groupon-style daily deals website into one of China’s dominant “super apps,” boasting over 770 million annual transacting users and 14.5 million merchants across food delivery, local retail, and hotel bookings.
However, facing intense domestic market competition and severe margin compression in its core logistics business, Meituan publicly committed to investing “billions” into artificial intelligence and domestic chip capabilities. This pivot began materializing in late 2025 with the release of the 560-billion-parameter LongCat-Flash model.
By open-sourcing a 1.6-trillion-parameter model under an MIT license, Meituan is signaling its ambition to become a foundational player in global AI infrastructure. The MIT license allows corporate engineering teams to deeply modify, compile, and hard-code LongCat-2.0 directly into closed-source commercial applications without the legal obligation to open-source their derivative works.
Capitalizing on U.S. Export Controls
Meituan’s breakthrough arrives at a critical juncture in the geopolitical tech war. Following requests from the U.S. government, OpenAI was recently forced to limit access to its new GPT-5.6 models, while Anthropic took its latest Claude Fable 5 and Mythos 5 models entirely offline. As EastFrontier reported, Chinese models stepped in to fill the gap left by the Fable 5 ban, a dynamic that LongCat-2.0 is now positioned to accelerate.
By locking down Western closed-source models and driving up API costs, these defensive regulatory maneuvers have inadvertently created a massive operational window for global developers seeking affordable, high-performance alternatives. With LongCat-2.0, Meituan has proven that China not only has the domestic silicon to train frontier-class models but also the strategic intent to open-source them and capture the global developer ecosystem that Western labs are being forced to abandon.