Qwen3.5 Plus

Baidu Baike documents that Qwen3.5 was launched on February 16, 2026, comprising Qwen3.5-Plus and Qwen3.5-397B-A17B versions, with open-source releases including a 2B dense model and a 35B-A3B MoE model. The architecture combines a Hybrid Attention Mechanism with a Sparse Mixture of Experts (MoE), where Qwen3.5-397B-A1 Qwen3.5-Plus achieves a score of 87.8 on the MMLU-Pro knowledge reasoning benchmark, with visual capabilities covering multimodal reasoning, text recognition, and video understanding, and the API price is reported at 0.8 RMB per million tokens. The model set new records for open-source models on multiple authoritative

The InferenceX profile of Qwen3.5-397B-A17B details the architecture and positioning of Alibaba's first open-weights Qwen3.5 release, which underpins the hosted Qwen3.5-Plus variant. It reports 397B total parameters with 17B activated per forward pass, a 1M-token default context for the hosted Plus service with built-i It documents the Qwen3.5 release timeline (official blog dated February 2026, public release mid-February 2026 — February 16 via OpenRouter, February 17 via Artificial Analysis) and the vendor's agentic, multimodal framing targeting reasoning, coding, agent capabilities, image and video understanding, and GUI interacti

The Qwen3.5 native vision-language series Plus models are built on a hybrid architecture that integrates linear attention mechanisms with sparse mixture-of-experts models, achieving higher inference efficiency. $0.26 per million input tokens, $1.56 per million output tokens. 1,000,000 token context window, maximum outp

Alibaba Unveils Qwen3.5-Plus, Undercutting Gemini 3 Pro on Cost - Chinese tech giant says new open-source AI model rivals Google’s flagship system while charging a fraction of the API price