iFlytek launches Spark X2-Flash, a large language model based on the MoE architecture.

iFlytek's Starfire X2-Flash has been officially launched and its API is now open. The model adopts the MoE architecture, with a total of 30 bytes of parameters, supports a 256K ultra-long context, and is trained on a Huawei Ascend 910B domestic computing power cluster. It significantly improves agent and code capabilities, with actual testing showing performance approaching that of a trillion-parameter model, while token consumption costs are less than one-third of mainstream large-scale models. The model has been integrated with platforms such as AstronClaw and Loomy, and is compatible with mainstream agent frameworks such as OpenClaw.