LongCat API Platform Change Log

Version: 2026-02-05

LongCat-Flash-Lite Release

  • LongCat-Flash-Lite is officially released. Built on a highly efficient MoE architecture (68.5B total parameters, about 3B activated per inference), the model achieves superior parameter utilization through an N-gram embedding table, specifically optimized for inference efficiency and specialized use cases.
  • Compared to models of similar scale, the core features are as follows:
    • Superior Inference Efficiency: By mitigating I/O bottlenecks within MoE layers via the N-gram embedding table—combined with specialized cache and kernel optimizations—the model significantly reduces latency and boosts overall efficiency.
    • Strong Agentic and Coding Performance: Demonstrates exceptional competitiveness in tool-use and coding proficiency, delivering high-tier performance relative to its model scale.
  • For calling instructions, please refer to: LongCat API Platform Interface Documentation
  • Open-Source Platforms:

Version: 2026-01-14

LongCat-Flash-Thinking-2601 Release

  • LongCat-Flash-Thinking-2601 is officially released. As an upgraded reasoning model, it is built on a Mixture-of-Experts (MoE) architecture with a total of 560 billion parameters. While maintaining strong competitiveness in traditional reasoning benchmarks, the model systematically enhances AI Agent reasoning capabilities through large-scale multi-environment reinforcement learning.

  • Compared to the LongCat-Flash-Thinking model, the core features of this upgrade are as follows:

    • Extreme Robustness in Noisy Environments: Through systematic curriculum training focused on noise and uncertainty in real-world environments, the model delivers exceptional performance in agent tool calling, agent search, and tool-fusion reasoning, with a significant boost in generalization capabilities.
    • Powerful Agent Capabilities: By constructing environment dependency maps with over 60 types of tools, combined with multi-environment expansion and large-scale exploration training, the model significantly improves generalization in complex, out-of-distribution (OOD) real-world scenarios.
    • Advanced Deep Thinking Mode: Utilizing parallel reasoning to expand thinking breadth, combined with a summarization mechanism featuring recursive feedback to expand thinking depth, effectively solving high-difficulty problems.

  • For calling instructions, please refer to: LongCat API Platform Interface Documentation

  • Open-Source Platforms:

Version: 2025-12-22

LongCat-Flash-Chat Update

  • The LongCat-Flash-Chat model has been upgraded to a new version. This update involves model capability enhancements only. The model name and API invocation method remain unchanged.

  • Building upon its principles of “extreme efficiency” and “lightning-fast responsiveness,” the new version of LongCat-Flash-Chat further improves its context handling and programming capabilities:

    • Significantly enhanced programming performance: Deep optimizations for developer-oriented use cases, with substantial improvements in code generation, debugging, and code explanation tasks. Developers are encouraged to prioritize evaluation and testing.
    • Support for a 256K ultra-long context length: The context length has doubled compared to the previous generation (128K), enabling efficient processing of large documents and long-sequence tasks.
    • Comprehensively strengthened multilingual capabilities: High-quality support for nine languages, including Spanish, French, Arabic, Portuguese, Russian, and Indonesian.
    • More robust Agent capabilities: Improved stability and efficiency in complex tool invocation and multi-step task execution.

  • For API usage details, please refer to the LongCat API Platform Interface Documentation.

Version: 2025-09-22

LongCat-Flash-Thinking Release

Version: 2025-09-05

LongCat API Platform Launch

  • The LongCat API Platform is now live, supporting API calls to the LongCat-Flash-Chat model. You can call it by specifying model=LongCat-Flash-Chat.
  • For calling instructions, please refer to: LongCat API Platform Interface Documentation

Version: 2025-08-29

LongCat-Flash-Chat Release

Last Updated:
Contributors: zhuqi09, wb_changzhe02