Meituan Ships LongCat-2.5-Preview: 1.6T MoE Multimodal Agent Model With 1M Context, Pandaily reports
Reported by Pandaily and IT Home · not yet confirmed by the company or a second independent outlet. We update this page when it is.
Meituan launched LongCat-2.5-Preview on September 25, 2026: a 1.6T MoE (~48B active) with native 1M-token multimodal context aimed at long-horizon software agents, without publishing benchmarks this round, Pandaily reported.
China context
- Original name
- 美团
- Outside China
- Not stated in the sources yet
- Claims
- Company-reported; not yet independently evaluated
- For builders
- Developers can evaluate LongCat-2.5-Preview for long-context agent tasks, but must confirm API access and pricing outside China.
- For investors
- Meituan's entry into frontier-scale MoE models signals increased competition in the Chinese AI API market, potentially pressuring pricing and differentiation.
Translated from Chinese. Quotes and facts link to the original sources.
Meituan's LongCat API platform released LongCat-2.5-Preview on September 25, 2026. The model uses a Mixture-of-Experts architecture with approximately 1.6 trillion total parameters and about 48 billion active parameters per inference. It natively supports a 1 million token context window and adds image understanding for multimodal tasks. The release emphasizes long-horizon agentic tasks, coding, and compatibility with developer environments such as Claude Code, Hermes, OpenClaw, OpenCode, and Kilo Code. API and web experience access are available. No benchmark results were published.
LongCat-2.5-Preview is a sparse Mixture-of-Experts model with ~1.6T total parameters and ~48B active parameters, enabling efficient inference for long-context tasks. It natively handles 1M tokens, suitable for large codebases, logs, and multi-turn agent workflows. Multimodal input includes image understanding for cross-modal QA, summarization, and visual reasoning. The model is positioned for agentic coding and integrates with popular developer tools.
Developers outside China gain a new API option for long-context multimodal agent tasks, but must verify API availability and pricing outside China before adoption.
LongCat-2.5-Preview targets enterprises and developers needing long-context multimodal agents for code understanding, automation, and complex document processing. Its 1M context and coding integrations could reduce the need for chunking or external retrieval in agent workflows, potentially lowering development cost and latency for long-horizon tasks.
Next observable signals: whether Meituan publishes benchmark results for LongCat-2.5-Preview, and whether the API becomes accessible outside China with documented pricing.