AI Model and Product Knowledge
This domain tests whether you can match Alibaba Cloud's model families and products to the right job. Know the Qwen (Tongyi Qianwen) family shape: general chat models, larger and smaller parameter tiers, long-context variants, and specialized models for code, math, and vision-language input. Understand that Qwen-VL handles image-plus-text prompts, while text-only Qwen models reject images. Know the role of Tongyi Wanxiang for image generation, and speech products (recognition and synthesis) as separate services rather than chat endpoints.
Concept traps are common here. A question may describe a task requiring image understanding and offer a text-only model as a distractor; the correct answer is a multimodal model. Another trap is confusing model size with capability for a specific task: a smaller specialized model can beat a larger general one on code or math. Also watch for concurrency and context-window limits framed as model features rather than API parameters.
Rule of thumb: choose the smallest model that meets the modality and quality bar, then scale up only if evaluation fails. Modality first, task specialization second, size third.
Worked example: a team must extract tables from scanned invoices. The input is an image, so the answer must be a vision-language model; a long-context text model cannot read pixels regardless of context length. If the follow-up asks to summarize the extracted text cheaply at scale, a smaller general text model is appropriate.
Also know that model knowledge includes pricing and quota concepts at a high level: token-based billing, input versus output tokens, and that streaming affects perceived latency, not total cost.