Ling-2.6-flash

Ling-2.6-Flash is an open-source instruction-following language model developed by InclusionAI, designed for AI Agents, complex task execution, and efficient reasoning workflows.
The model focuses on faster responses, improved task execution, and reduced token consumption, making it suitable for practical AI applications and production environments.
Built with an efficient Mixture-of-Experts (MoE) architecture, Ling-2.6-Flash features approximately 104B total parameters with around 7.4B active parameters, delivering strong intelligence while reducing inference costs.
The model introduces a hybrid attention architecture that combines Lightning Attention and MLA (Multi-head Latent Attention), improving long-context processing efficiency and overall inference speed.
With support for up to 262K-token context windows, Ling-2.6-Flash can handle long documents, code analysis, knowledge tasks, tool calling, and automated AI Agent workflows.
Optimized for real-world Agent applications, the model delivers strong performance in tool usage, multi-step planning, and task execution. It can be integrated into AI coding workflows and development frameworks such as Claude Code, Kilo Code, and Qwen Code.