Meituan Open-Sources LongCat-2.0, a 1.6T Agentic Coding Model Built on Chinese Chips
The model, which spent two months atop OpenRouter under a stealth identity, is now publicly available under an MIT license with aggressive promotional pricing.
Chinese food-delivery and super-app company Meituan has officially released LongCat-2.0, a 1.6-trillion-parameter Mixture-of-Experts (MoE) model optimized for autonomous software engineering, publishing it on GitHub, Hugging Face, and its own platform under a commercially permissive MIT license. [1]
The release also revealed that LongCat-2.0 was the model behind “Owl Alpha,” an anonymous stealth model that had been running on the OpenRouter model-routing platform for approximately two months before Meituan claimed it. [1]
During that unbranded period, Owl Alpha processed roughly 10.1 trillion monthly tokens — averaging 559 billion tokens per day — representing a 242% month-over-month increase in volume that placed it among OpenRouter’s global top three models. [1] By the time Meituan identified itself as the developer, the model had reached the top ranking on the Hermes Agent workspace, second place on Claude Code deployments, and third place across international OpenClaw environments. [1]
Architecture and Context Window
LongCat-2.0 scales to 1.6 trillion total parameters while limiting active computation to an average of 48 billion parameters per token, with dynamic activation ranging from 33 billion to 56 billion parameters depending on query complexity. [1] The model supports a native 1-million-token context window. [1]
To sustain that context window without severe memory overhead, Meituan developed LongCat Sparse Attention (LSA), described as an evolution of DeepSeek Sparse Attention. [1] LSA addresses the quadratic memory costs typical of fine-grained sparse attention through three mechanisms: Streaming-aware Indexing, which restructures token selection into sequential memory blocks to improve High Bandwidth Memory (HBM) utilization; Cross-Layer Indexing, which amortizes attention calculation costs across adjacent model layers; and Hierarchical Indexing, which applies a two-stage coarse-to-fine scoring approach to filter token candidates before fine-grained selection. [1]
The architecture also incorporates an N-gram Embedding module that appends 135 billion parameters to a 5-gram token combination framework, expanding the core embedding space by roughly 100-fold to capture local token relationships and reduce memory input/output bottlenecks during large-batch inference. [1]
Post-Training and Benchmarks
LongCat-2.0’s post-training uses a framework Meituan calls Multi-Teacher Optimization via Mixture of Specialized Experts (MOPD), which separates optimization into three independent expert clusters: Agent Experts focused on tool invocation and self-correcting execution loops; Reasoning Experts optimized for multi-hop logic and STEM problem-solving; and Interaction Experts focused on instruction-following, factual grounding, and safety. [1] A dynamic gate-routing mechanism combines these behaviors at runtime. [1]
On the SWE-bench Pro software engineering benchmark, LongCat-2.0 scores 59.5, compared to 58.6 for OpenAI’s GPT-5.5. [1] The model also scores 70.8 on Terminal-Bench 2.1, 77.3 on SWE-bench Multilingual, and 73.2 on the general corporate workflow simulator FORTE. [1] The source notes that LongCat-2.0 trails premium frontier systems such as Claude Opus 4.8 on broader general-agent benchmarks like FORTE and BrowseComp. [1]
Pricing and Commercial Model
Standard pay-as-you-go API pricing is set at $0.75 per million input tokens and $2.95 per million output tokens. [1] A limited-time promotional discount reduces those rates to $0.30 per million input tokens and $1.20 per million output tokens. [1] All context-cache hits are processed at no charge under either pricing tier. [1]
Meituan also offers structured “Token Packs” — fixed volumetric allocations valid for 30 days — released via flash sales four times daily at 10:00, 16:00, 21:00, and 23:00 Beijing Time on a first-come, first-served basis. [1] The zero-cost cache-hit policy means that in long agentic sessions where a model repeatedly references the same large codebase, developers are only billed for cache misses and new output tokens. [1]
Trained Entirely on Domestic Chinese Chips
Meituan states that LongCat-2.0 was trained on a cluster of more than 50,000 domestic Chinese Application-Specific Integrated Circuits (ASICs), without relying on Nvidia GPUs. [1] The source frames this as evidence that near-frontier models can be scaled on non-U.S. silicon, though it also notes the broader context: the U.S. government has pressured American AI labs to restrict access to their latest models, with OpenAI limiting access to its GPT-5.6 models and Anthropic taking its Claude Fable 5 / Mythos 5 models entirely offline following government orders. [1]
Licensing and Company Background
The MIT license allows developers to modify, integrate, and redistribute LongCat-2.0 within closed-source commercial products without any obligation to disclose derivative work, in contrast to copyleft licenses such as the GNU General Public License (GPL). [1]
Meituan was founded in March 2010 by Wang Xing and reports more than 770 million annual transacting users and a merchant network of over 14.5 million. [1] The company trades on the Hong Kong Stock Exchange. [1] Its AI push began in late 2025 with LongCat-Flash, a 560-billion-parameter MoE model, followed by the reasoning-focused LongCat-Flash-Thinking. [1]
Sources
This article was drafted with AI from the cited sources and checked against them before publication. Spot an error? Let us know.
OpenAI's first custom silicon, co-developed with Broadcom, is designed from scratch for large language model inference and is targeted for gigawatt-scale deployment by late 2026.
OpenAI and Broadcom have jointly announced “Jalapeño,” a custom accelerator chip built specifically for large language model (LLM) inference — the process of running a trained AI model to generate outputs. [1] The chip is described as OpenAI’s first “Intelligence Processor” and the first in a planned multi-generation hardware platform the two companies are developing together. [1]
Broadcom CEO Hock Tan and President Charlie Kawwas presented the first wafer to OpenAI CEO Sam Altman and President Greg Brockman at the announcement. [1] For OpenAI, which has until now focused on model and product development rather than hardware, the announcement marks its first move into custom silicon. [1]
According to OpenAI, Jalapeño was not derived from a general-purpose chip but was designed from the ground up for modern LLM inference workloads. [1] The division of labor in the project has OpenAI leading chip design, Broadcom contributing silicon manufacturing and networking technology — including its Tomahawk networking chips — and Celestica handling boards, racks, and system integration. [1]
OpenAI says the chip’s architecture reduces data movement and pushes hardware utilization closer to its theoretical maximum. [1] Engineering samples are already running machine learning workloads in the lab, including the GPT-5.3-Codex-Spark model, which currently also runs on Cerebras hardware. [1] Early internal tests showed performance per watt that is “substantially better” than current state-of-the-art hardware, though OpenAI itself notes these figures have not been finalized. [1] The company says a technical report will follow, but at this stage the specific chips Jalapeño was benchmarked against, the tasks used, and the testing conditions have not been disclosed. [1]
OpenAI claims the design-to-tape-out process — the full cycle from initial design to a completed chip layout ready for manufacturing — took nine months, which the company describes as the fastest application-specific integrated circuit (ASIC) development cycle for high-performance semiconductors it is aware of. [1] OpenAI also says its own AI models helped accelerate parts of the chip design process. [1] Reports of OpenAI’s custom chip ambitions had been circulating since 2023. [1]
Broadcom CEO Tan has said the first deployment is planned for late 2026 at gigawatt scale, in partnership with Microsoft and other unnamed partners. [1] Broadcom has reportedly required Microsoft to guarantee it will purchase 40 percent of the chips in order to secure the first phase of production. [1] OpenAI frames the broader effort as evidence that controlling the full stack — from chip to product — enables faster, more reliable, and lower-cost model serving. [1]
Sources
This article was drafted with AI from the cited sources and checked against them before publication. Spot an error? Let us know.