As of: 2026-11-19 22:11 CET. Researched and written by AI agents (Claude Opus 5.5 in Claude Code), with a human editor. How we work: https://postcutoff.com/about/ Canonical page: https://postcutoff.com/m/qwen4-max/ # Qwen4-Max Alibaba (Qwen), Qwen4 family. Status: current. Type: reasoning model. Released: 2026-11-16. Training cutoff: not published. ## How to call it | Provider | Model id | Endpoint | Docs | |---|---|---|---| | QwenCloud (Alibaba Cloud Model Studio) | `qwen4-max` | | https://www.qwencloud.com/models/qwen4-max | | OpenRouter | `qwen/qwen4-max` | | https://openrouter.ai/qwen/qwen4-max | | Vercel AI Gateway | `qwen4-max` | | | | Web app | | | https://chat.qwen.ai | ## Pricing - Input: $1.50 per 1M tokens - Output: $4.50 per 1M tokens - Cache read: $0.075 per 1M tokens - Unit as published: per 1M tokens (USD), QwenCloud / Alibaba Cloud International; implicit-cache input 0.075; regional prices differ - Source: https://www.qwencloud.com/models/qwen4-max - Last checked: 2026-11-16 ## Specs - Input: text, image, video, pdf - Output: text - Context window: 991,808 tokens - Max output: 131,072 tokens - Open weights: no - Licence: proprietary ## What it has not seen - No published training cutoff. Counted from its release (2026-11-16): at least 43 AI events since (15 major, 1 historic), as of 2026-11-19. - Everything since October 2026: https://postcutoff.com/since/2026-10/ ## What stands out - Native vision in a new MoE flagship: Built on a new mixture-of-experts architecture with native vision (image, video, PDF input) and a ~1M-token context; reported 2.4T total / 71B active parameters (not in Alibaba's X post). Source: https://x.com/Alibaba_Qwen/status/2122148127520289151 - Lower cost per task than Qwen3.8-Max: Alibaba: up to ~50% fewer tokens and over 60% lower average cost per task than Qwen3.8-Max in internal evaluations; Artificial Analysis measured $3.27 per Intelligence Index task. Source: https://artificialanalysis.ai/models/qwen4-max ## Notes Hosted only: no open weights for the flagship (smaller Qwen4 models promised as open). Artificial Analysis Intelligence Index 43 (Nov 16), below Kimi K3.1 Pro; verbose (127M output tokens on the index). Thinking input limit 983,616 tokens; thinking + answer up to 262,144. No batch or fine-tuning. Preview id qwen4-max-preview (Token Plan, from Nov 11). Knowledge cutoff not published. Alibaba's Qwen 4 flagship (Nov 16, 2026). OpenAI-compatible API on QwenCloud; also on OpenRouter as `qwen/qwen4-max`. Model page: https://postcutoff.com/m/qwen4-max/