Ling-2.6-Flash
About
InclusionAI's efficient 104B MoE instruct model with only 7.4B active parameters per token. Purpose-built for agentic workflows requiring fast responses and high token efficiency. Achieves 59.3% on GPQA Diamond. Nearly double the Artificial Analysis Intelligence Index score of comparable open-weight models. Available free on OpenRouter (inclusionai/ling-2.6-flash:free).
Capabilities
VisionMultimodalReasoningFunction CallingTool UseStructured OutputsCode Execution
Specifications
FamilyLing 2.6
Released2026-04-21
Parameters104B (7.4B activated)
Context262K
Architecturemoe
Specializationgeneral
LicenseApache 2.0
Trainingpretrained
Created by
InclusionAI is Ant Group's artificial general intelligence research lab, responsible for developing
Hangzhou, China
Founded 2023
Website