Qwen3 Next 80B A3B

Qwen3 Next 80B A3B is a large language model released by Alibaba as part of the Qwen3 line. Its name follows the Qwen convention for sparse mixture-of-experts (MoE) models: roughly 80 billion total parameters with approximately 3 billion active parameters per token (the "A3B" suffix). This design aims to give the model the knowledge capacity associated with a large parameter count while keeping the per-token compute closer to that of a much smaller dense model, which is the main reason it is positioned as an efficiency-oriented member of the Qwen3 family rather than a maximum-capability endpoint. In measured serving performance, Qwen3 Next 80B A3B reaches around 185.6 output tokens per second with a time to first token of roughly 992 ms, according to Artificial Analysis. That throughput figure reflects the low active-parameter count: fewer parameters are engaged per generated token, so hosted deployments can produce output quickly relative to dense models of comparable total size. The sub-second time to first token makes it usable in interactive settings, though prompt length and provider infrastructure will move both numbers in practice. We do not track a confirmed context window, modality list, or benchmark suite for this entry, so treat capability claims beyond the throughput measurements above as unverified here — check the provider's own model card for the variant you intend to use, since Qwen models are frequently published in separate instruct-tuned and reasoning-oriented builds with different behavior. In practice, models with this shape are deployed where high token throughput per unit of serving cost matters more than squeezing out the last few points of benchmark accuracy: long-form generation, batch text processing, and agent loops that emit many tokens per task.