Alibaba’s Qwen3.8‑Max is now available via a hosted API, but the 2.4‑trillion‑parameter mixture‑of‑experts model cannot be run on a single workstation. The open weights will be released next week, yet Alibaba has not disclosed the activated‑parameter count, making it impossible to estimate serving cost or hardware needs for self‑hosting. For most teams the immediate, low‑risk path is to call the model through the DashScope or OpenAI‑compatible endpoint, changing only the base URL and model ID. Pricing is $2 per million input tokens, $6 per million output tokens, with an implicit cache read at $0.25 per million tokens; cached input is eight times cheaper than fresh input, so stable prefixes save money. The model accepts text, image and video and returns text, with a 1‑million‑token context window (991 K max input, 131 K max output, 262 K reasoning budget) and rate limits of 2 M tokens per minute and 15 K requests per minute. Built‑in tools include code_interpreter, web_search, web_extractor, t2i_search and i2i_search, enabling repository‑scale coding agents, long‑document knowledge bases, video indexing and structured data extraction without extra integration work. If you need to keep data on‑premise or handle regulated information, wait for the Qwen3.8‑27B checkpoint, which fits ordinary GPU hardware and will also be released with open weights next week. Use the hosted flagship for exploratory agent work and deploy the 27B model on your own servers for PII‑bearing or high‑volume tasks. Once the license file and activated‑parameter numbers are published, evaluate a multi‑node cluster for the full 2.4T weights. Until then, the API gives you immediate access, predictable cost, and a clear upgrade path.
#AI #Product #LLM #MachineLearning #GenAI #CloudAPI