The Qwen3.5 series 397B-A17B native vision-language model is built on a hybrid architecture that integrates a linear attention mechanism with a sparse mixture-of-experts model, achieving higher inference efficiency. It delivers state-of-the-art performance comparable to leading-edge models across a wide range of tasks, including language understanding, logical reasoning, code generation, agent-based tasks, image understanding, video understanding, and graphical user interface (GUI) interactions. With its robust code-generation and agent capabilities, the model exhibits strong generalization across diverse agent.
Evaluations
13
across 13 benchmarks
Latency
4m 13s
Context length
262k
tokens
Cost
$0.60 · $3.60
input · output per 1M tokens
Key takeaways
Qwen3.5 397B A17B is a multimodal vision-language model designed for diverse tasks, including language, image, and video understanding, logical reasoning, code generation, and GUI interactions.
The model utilizes a hybrid architecture combining linear attention with a sparse Mixture-of-Experts (MoE) to achieve higher inference efficiency and state-of-the-art performance across various domains.
The model demonstrates exceptional problem-solving capabilities in advanced mathematics, achieving 100% accuracy on the AIME 2024 benchmark, and shows proficiency in addressing logic-based errors in symbolic manipulation.
A significant limitation is its struggle with complex symbolic evaluation and parsing, particularly with LaTeX expressions and handling symbolic 'i' in complex expressions, resulting in a 20% score on SWE-bench Lite.
Key weaknesses include difficulties with operator precedence, grouping, and consistently handling parsing in 'evaluate=False' contexts, along with issues in unit prefix multiplication and dimensional analysis.
