Four lightweight models of the Qwen 3.5 series will be released at once, an open model with higher performance than the GPT-5 mini

Alibaba's AI research team Qwen (Tongyi Lab) announced four medium models of the Qwen3.5 series on February 25, 2026: Qwen3.5-122B-A10B , Qwen3.5-27B , Qwen3.5-35B-A3B , and Qwen3.5-Flash .

🚀 Introducing the Qwen 3.5 Medium Model Series
Qwen3.5-Flash · Qwen3.5-35B-A3B · Qwen3.5-122B-A10B · Qwen3.5-27B

✨ More intelligence, less computation.
• Qwen3.5-35B-A3B now surpasses Qwen3-235B-A22B-2507 and Qwen3-VL-235B-A22B — a reminder that better architecture, data quality,… pic.twitter.com/ZWPibMn6at

— Qwen (@Alibaba_Qwen) February 24, 2026

1/3 Qwen3.5 Open-Source Family Expands 🚀
More intelligence, less compute. Introducing the Medium Model Series: Qwen3.5-Flash, Qwen3.5-35B-A3B, Qwen3.5-122B-A10B, Qwen3.5-27B
Qwen3.5-35B-A3B: Surpasses the previous Qwen3-235B-A22B-2507 and Qwen3-VL-235B-A22B. Proof that… pic.twitter.com/OLYfFW762w

— Tongyi Lab (@Ali_TongyiLab) February 25, 2026

The Qwen3.5 series is an AI model series that was released on February 16, 2026. The first product, the Qwen3.5-397B-A17B, was released as an open model at the same time as its announcement. While the Qwen3.5-397B-A17B is a free, downloadable open model, it offers performance comparable to closed models such as GPT-5.2 and Claude Opus 4.5.

Chinese AI 'Qwen3.5-397B-A17B' equivalent to GPT-5.2 appears as an open model - GIGAZINE

Of the four models announced this time, three are open-source models: the Qwen3.5-122B-A10B, the Qwen3.5-27B, and the Qwen3.5-35B-A3B. The Qwen3.5-122B-A10B boasts performance exceeding that of the previous generation heavyweight model, the Qwen3-235B-A22B-2507. The Qwen3.5-27B and Qwen3.5-35B-A3B are also advertised as 'bridging the gap between mid-range and cutting-edge models.'

Below is a graph showing the benchmark results for 'Qwen3.5-122B-A10B', 'Qwen3.5-27B', 'Qwen3.5-35B-A3B', 'Qwen3-235B-A22B-2507', 'GPT-5 mini', and 'gpt-oss-120b' from left to right. It can be seen that Qwen3.5-122B-A10B and Qwen3.5-27B perform better than GPT-5 mini.

The graph below shows the results of tests measuring agent performance, sorted in order of Qwen3.5-122B-A10B, Qwen3.5-27B, Qwen3.5-35B-A3B, GPT-5 mini, and gpt-oss-120b. The Qwen3.5 series is characterized by its high agent performance, and even Qwen3.5-35B-A3B, which has 3 billion active parameters, outperforms GPT-5 mini.

The Qwen3.5 series also boasts high visual comprehension performance. Comparing the visual comprehension benchmark scores of 'Qwen3.5-122B-A10B,' 'Qwen3.5-27B,' 'Qwen3.5-35B-A3B,' 'Qwen3-VL-235B-A22B,' 'GPT-5 mini,' and 'gpt-oss-120b,' the high performance of the Qwen3.5 series stands out.

Model data for 'Qwen3.5-122B-A10B', 'Qwen3.5-27B', and 'Qwen3.5-35B-A3B' can be downloaded from the links below.

Qwen3.5 - a Qwen Collection
https://huggingface.co/collections/Qwen/qwen35

Qwen3.5-Flash is a production version developed based on Qwen3.5-35B-A3B and is available via Alibaba Cloud's API. The usage fee per million tokens is $0.1 (approximately ¥15.6) for input and $0.4 (approximately ¥62.4) for output.

ModelStudio Console
https://modelstudio.console.alibabacloud.com/ap-southeast-1/?tab=doc#/doc/?type=model&url=2840914_2&modelId=group-qwen3.5-flash