Tuesday, June 30, 2026

Today’s Edition

AI Intel Report

MARKETS

Frontier Models

Meituan LongCat-2.0 1.6T MoE Model Edges GPT-5.5 on SWE-bench Pro

The open-sourced release from the Chinese delivery firm features 1 million token context and ASIC-based training while powering agent capabilities on OpenRouter.

3 MIN READ
Inside a large industrial logistics and technology facility operated by a Chinese delivery company rows of tall black server racks fill a brightly lit warehouse style hall each rack densely packed with custom ASIC accelerator boards connected by thick bundles of colored network cables and liquid cooling pipes running along the floor and ceiling the hardware is actively engaged in distributed training runs for a large scale mixture of experts artificial intelligence model with indicator lights blinking in steady patterns on every board in the background through open loading bay doors electric delivery scooters and cargo vans are parked in neat lines being loaded with packages by workers wearing reflective vests while additional racks of identical ASIC hardware stand ready for scaling the one million token context capability of the model the overall environment mixes high density compute infrastructure with everyday delivery operations including conveyor belts carrying parcels and charging stations for fleet vehicles creating a unified scene that shows the integration of advanced model training hardware within an operational delivery firm setting where the same organization develops and deploys the LongCat-2.0 system alongside its logistics network the floor is concrete with yellow safety lines the ceiling has exposed ductwork and fluorescent panels the air appears cool and filtered with faint mist from cooling units the composition centers on the hardware racks as the dominant element while the delivery vehicles and workers provide contextual depth without any close up faces or direct interaction the entire view is captured from a medium distance showing depth through multiple parallel aisles of equipment and vehicles extending into the distance emphasizing the scale of both the compute cluster and the physical logistics operations that the model supports through agent capabilities routed via external platforms
Illustration: AI Intel Report

LongCat-2.0 is a 1.6 trillion parameter Mixture-of-Experts language model from Meituan designed specifically for agentic coding tasks.

Meituan released LongCat-2.0 as an open source model.

The model targets agentic coding applications.

What are the core specifications of LongCat-2.0?

LongCat-2.0 maintains 1.6 trillion total parameters.

Approximately 48 billion parameters activate per token.

The model supports a native 1 million token context window.

LongCat-2.0 incorporates LongCat Sparse Attention.

N-gram Embedding modules form part of the architecture.

How was LongCat-2.0 trained?

Both the full training run and large-scale deployment rely on AI ASIC superpods.

Pretraining spans millions of accelerator-days.

The process covers more than 35 trillion tokens.

Training involved more than 50K AI ASICs.

No rollbacks or irrecoverable loss spikes occurred.

What benchmarks does LongCat-2.0 achieve?

LongCat-2.0 scores 59.5 on SWE-bench Pro.

GPT-5.5 scores 58.6 on the same benchmark.

LongCat-2.0 scores 70.8 on Terminal-Bench 2.1.

We are introducing and open sourcing LongCat-2.0, a large-scale MoE language model with 1.6 trillion total parameters and ~48 billion activated per token — a substantial step up from previous LongCat models, accompanied by several architectural improvements.LongCat team

What specialized components support agentic coding?

Specialized experts form distinct groups.

One expert group handles Agent tasks.

A separate group manages Reasoning.

Another expert group addresses Interaction.

How does LongCat-2.0 connect to platforms?

LongCat-2.0 serves as the model behind Owl Alpha on OpenRouter.

Open sourcing enables community access to the full model.

LongCat-2.0 Model Specifications and Performance
SpecificationValue
Total Parameters1.6 trillion
Active Parameters~48 billion
Context Window1 million tokens
Pretraining Tokens>35 trillion
Training HardwareAI ASIC superpods
Key BenchmarksSWE-bench Pro: 59.5, Terminal-Bench 2.1: 70.8

What steps allow users to engage with LongCat-2.0?

  1. Access the model weights from the Hugging Face repository.
  2. Review the full announcement and technical details on the LongCat blog.
  3. Evaluate performance through SWE-bench Pro and Terminal-Bench 2.1.
  4. Integrate the model via OpenRouter for Owl Alpha agent use.
  5. Experiment with LongCat Sparse Attention in custom agent workflows.

What implications follow for stakeholders?

Training on domestic AI ASIC hardware demonstrates scalable infrastructure.

Open sourcing at this scale adds to available frontier models.

Agentic coding optimization targets specific enterprise and developer needs.

What developments could come next?

Additional expert specializations may expand capabilities.

Further refinements to sparse attention could improve efficiency.

New fine-tunes or tools may build on the released base model.

Frequently asked

What benchmark scores does LongCat-2.0 report?

LongCat-2.0 achieves 59.5 on SWE-bench Pro and 70.8 on Terminal-Bench 2.1. These figures exceed the GPT-5.5 result of 58.6 on SWE-bench Pro.