Qwen3.6 35B A3B

Qwen3.6 35B A3B is a language model released by Alibaba as part of the Qwen3.6 series. Its name follows the convention Alibaba uses across recent Qwen generations: a total parameter count (35B) paired with an "A3B" suffix denoting the number of active parameters per token (approximately 3B). That structure indicates a mixture-of-experts architecture, where only a fraction of the network is engaged for any given token, rather than a dense model of the same nominal size. In measured serving benchmarks collected by Artificial Analysis, this model produced roughly 143 output tokens per second with a time to first token of about 1,077 ms. The high throughput figure is consistent with the sparse-activation design, since compute per token scales with active parameters rather than total parameters. The roughly one-second first-token latency is a separate axis and depends heavily on the specific provider, batching policy and prompt length, so it can differ from the figure above across the endpoints listed on this page. We do not currently track a confirmed context window, modality list or benchmark suite for Qwen3.6 35B A3B in our database, so those details should be verified against the provider you plan to use. In practice, models with this shape are typically selected when throughput per unit of compute matters more than raw parameter count — high-volume text generation, chat backends and agent loops that issue many short calls. Compare the providers in the pricing table above, since serving configuration for MoE models varies noticeably between hosts.