Alibaba: qwen3-next-80b-a3b-thinking API

The flagship deep reasoning model of the Qwen3-Next series, released by Alibaba Cloud on September 11, 2025. It features a pioneering Hybrid Transformer-Mamba and High-Sparsity Mixture-of-Experts (MoE) architecture. With 80 billion (80B) total parameters, it activates only 3 billion (3B) parameters per token during inference (A3B). Operating natively in Thinking Mode, it automatically generates structured <think> traces for complex logic. This model represents the ultimate balance between "extreme inference efficiency" and "flagship-level cognitive depth." It delivers a 10x throughput improvement over previous 32B dense models while achieving a staggering 87.8 score on AIME 2025, bridging the intelligence gap between sparse MoE and trillion-parameter dense flagships.

Frequently Asked Questions

What is the context window of qwen3-next-80b-a3b-thinking?

qwen3-next-80b-a3b-thinking supports a context window of up to 131,072 tokens.

Does qwen3-next-80b-a3b-thinking support function calling?

Yes. qwen3-next-80b-a3b-thinking supports tool / function calling.

Does qwen3-next-80b-a3b-thinking support reasoning?

Yes. qwen3-next-80b-a3b-thinking is a reasoning-capable model.

What is the knowledge cutoff of qwen3-next-80b-a3b-thinking?

The knowledge cutoff of qwen3-next-80b-a3b-thinking is 2025-03-31.