LongCat 2.0 by LongCat is a 1.6T MoE model with 1M context, LongCat Sparse Attention, N-gram embeddings, and MOPD multi-expert fusion for agentic coding.
Capabilities, design details, and architectural traits
LongCat 2.0 is a 1.6-trillion-parameter MoE language model with ~48 billion parameters activated per token, built specifically for agentic coding and long-context agent tasks. Its entire training and deployment run on AI ASIC superpods - a 50,000-card domestic compute cluster.
Before its official reveal, the model operated anonymously as Owl Alpha on OpenRouter for roughly two months, consuming ~10.1 trillion tokens in a single month.
| Trait | Detail |
|---|---|
| LongCat Sparse Attention (LSA) | Evolution of DeepSeek Sparse Attention with a lighter indexer, providing linear-complexity attention for 1M-token context |
| N-gram Embedding module | Expands the embedding space by ~100x through N-gram token combinations, capturing richer local context |
| Zero-computation experts + ScMoE | Token-level dynamic compute allocation with 33B-56B activated per token |
| MOPD multi-expert fusion | Agent, Reasoning, and Interaction experts dynamically routed per task |
| 1M-context training | Trained on hundreds of billions of tokens of 1M-context data with dedicated post-training |
| ASIC superpod training | Full pretraining across 35+ trillion tokens on AI ASIC hardware with no rollbacks or irrecoverable loss spikes |
LongCat 2.0 is deeply integrated with mainstream agentic harnesses including Claude Code, OpenClaw, Hermes, OpenCode, and Kilo Code, targeting repository-level edits, automated task execution, and multi-step agentic workflows.
Independent evaluations · Artificial Analysis
Evaluate specifications, pricing, and independent benchmark indices
| Model Details | |||
|---|---|---|---|
| General Info | |||
| Provider | LongCat | Anthropic | Anthropic |
| Release Date | June 29, 2026 | July 24, 2026 | June 9, 2026 |
| Knowledge Cutoff | - | May 2026 | - |
| Context & Limits | |||
| Context Window | 1M | 1M | 1M |
| Pricing (per 1M tokens) | |||
| Input Pricing | $0.75 Best Input Pricing | $5 | $10 |
| Output Pricing | $2.95 Best Output Pricing | $25 | $50 |
| Modalities | |||
| Inputs | text | textimage | textimagefile |
| Outputs | text | text | text |
| Benchmarks (0-100) | |||
| Intelligence Index | 34.0 | 63.1 Best Intelligence Index | 62.1 |
| Coding Index | 45.3 | 78.0 Best Coding Index | 76.5 |
| Agentic Index | 22.0 | 59.2 Best Agentic Index | 56.6 |
Graduate-level reasoning and expert Q&A evaluation.
Extremely difficult logical reasoning and knowledge.
Logical reasoning over long context windows.
Independent evaluation data provided by Artificial Analysis. To view the latest benchmarks and full details, visit their official site.
