Qwen3.5 9B (DeepInfra)

DeepInfra hosts a model labeled Qwen3.5-9B under the Qwen namespace, presented as a member of Alibaba's Qwen3.5 series and positioned as a high-performance text model in that lineup. The Qwen3.5 9B API Benchmarks post on DeepInfra's blog treats this model as its central subject, publishing a dedicated latency, throughput, and cost study for the same checkpoint, which signals that DeepInfra is actively benchmarking the deployment behavior of this 9B model rather than just listing it.

A DeepInfra-authored overview characterizes Qwen3.5-9B's architecture as a hybrid design that combines Gated Delta Networks, a form of linear attention, with a sparse Mixture-of-Experts system, aimed at balancing long-context linear attention with conditional expert routing. The same page positions it as a flagship of the Qwen3.5 Small Model Series, suggesting the 9B variant is intended as a capable yet compact text model suitable for general language workloads routed through DeepInfra's serving stack.