Recently, the Qwen team announced that their open-source large model family Qwen3.5 has achieved significant expansion, with multiple new models released at once and corresponding production-grade API services launched simultaneously.

The newly added open-source models mainly include three:

  1. Qwen3.5-122B-A10B: This model performs outstandingly in complex agent tasks (such as multi-step reasoning and tool calls), further narrowing the performance gap between small and medium models and top closed-source models.

  2. Qwen3.5-35B-A3B: Its performance has fully surpassed previous larger parameter scale models such as Qwen3-235B-A22B, demonstrating a technical approach that improves intelligence through optimized architecture, data, and reinforcement learning collaboration, rather than simply increasing parameters.

  3. Qwen3.5-27B (Dense): A model that focuses on "small size and high efficiency," aiming to lower the entry barrier for using large models.

At the same time, for enterprise-level production applications, the Alibaba Cloud BaiLian platform has officially launched the Qwen3.5-Flash API. This service is a managed version aligned with Qwen3.5-35B-A3B, defaulting to support a context length of up to 1 million tokens and including an official native toolchain, ready to use without additional integration.

QQ20260225-092635.png