{"id":"qwen4-max","slug":"qwen4-max","name":"Qwen4-Max","orgId":"alibaba","family":"qwen4-max","releaseDate":null,"availability":"ga","openWeights":false,"license":"Not stated","paramsTotal":2400000000000,"paramsSource":"tracked","contextWindow":null,"knowledgeCutoff":null,"modelCardUrl":"https://qwen.ai/blog?id=qwen4-max","hfRepo":null,"modalities":["text","image","video"],"aliases":["Qwen4-Max","qwen4-max","Qwen4 Max","qwen 4 max","qwen4.0-max","Qwen4.0-Max","Qwen4.0 Max"],"aliasNote":null,"origin":"tracked","tracked":{"trackedSince":"2026-11-16","announced":"2026-11-16","blurb":"Qwen's production multimodal model in the Qwen4 line, a separate API model from Qwen4-Max-Preview: text generation, image and video understanding, and deep reasoning, with a 1M-token context. Official Alibaba sources describe a sparse MoE with 2.4 trillion total parameters and 71 billion active per token. Announced November 16, 2026 (official Qwen and Alibaba Cloud blog and changelog; @Alibaba_Qwen at 2026-11-16T02:01:55Z, which is 9:01 PM ET on November 15). Live on QwenCloud as a self-serve pay-as-you-go API and Token Plan (model qwen4-max), and in Qwen Studio, QwenWork, and Qoder. No public weights for this model; Qwen says smaller Qwen4 models will get open weights later.","api":"live","apiModelId":"qwen4-max","openWeights":"no","params":{"text":"2.4T total, 71B active (MoE)","source":"Alibaba Cloud Community blog","total":2400000000000},"facts":[{"label":"Type","value":"Multimodal model for text generation, image and video understanding, and deep reasoning. Thinking is on by default; reasoning_effort can be set to low, high, or xhigh (Alibaba Cloud Model Studio and QwenCloud docs)"},{"label":"Architecture","value":"Sparse Mixture-of-Experts; official Alibaba blog states 2.4 trillion total parameters and 71 billion active per token, 25% fewer active parameters than Qwen3.8-Max at the same 2.4T scale, with changes across attention, residual connections, and embeddings"},{"label":"Context","value":"1,000,000-token context. Official QwenCloud and Model Studio docs: maximum output 131,072 tokens without thinking; maximum thinking plus final answer 262,144 tokens; maximum input 991,808 tokens (non-thinking) or 983,616 tokens (thinking)"},{"label":"Tools","value":"Function calling, web search, structured outputs, explicit and implicit caching, chain-of-thought summary, PDF understanding, code interpreter, web scraping, and MCP. Compatible with OpenAI Chat Completions, Responses, Anthropic Messages, and DashScope (QwenCloud and Model Studio docs)"},{"label":"Hosted access","value":"Self-serve QwenCloud pay-as-you-go API and Token Plan, model qwen4-max. Singapore Chat Completions base URL https://ap-southeast-1.maas.qwencloudapi.com/compatible-mode/v1. Also listed on Alibaba Cloud Model Studio and in Qwen Studio, QwenWork, and Qoder. The separate qwen4-max-preview model remains Token Plan only through 24:00 November 25, 2026 (UTC+8)"},{"label":"Price","value":"Singapore / International: $1.50 per 1M input tokens and $4.50 per 1M output tokens, with implicit cache hits at $0.075 per 1M tokens (Alibaba Cloud Model Studio pricing, official Qwen post, and Alibaba Cloud Community blog). Official Model Studio tables list $1.413 input and $4.24 output per 1M tokens in Beijing and other Global regions. Pay-as-you-go calls in Singapore include 1 million free tokens, separate from Token Plan credits (QwenCloud docs)"},{"label":"Weights","value":"No public weight release for Qwen4-Max. Official announcement says open weights of smaller Qwen4 models will be released later"},{"label":"License","value":"Not stated"},{"label":"Related","value":"Separate from Qwen4-Max-Preview (model qwen4-max-preview) and from earlier Qwen Max, Qwen3 Max, Qwen3.8 Max, and Qwen VL Max pages"}],"claims":["Lab claim: on Alibaba's internal Work360 benchmark (721 specialized fields), Qwen4-Max used 35% fewer tokens than Qwen3.8-Max at High effort reasoning mode and reduced average cost per task by around 60% (self-reported, Alibaba Cloud Community blog).","Lab claim: official @Alibaba_Qwen announcement says Qwen4-Max uses up to about 50% fewer tokens than Qwen3.8-Max, with average cost per task down over 60% in Qwen's internal evaluation (self-reported)."],"pitch":"a multimodal model with a 1M-token context, native vision, and a live QwenCloud API (model qwen4-max)","sources":[{"kind":"lab","title":"Qwen blog: Qwen4-Max","url":"https://qwen.ai/blog?id=qwen4-max","date":"2026-11-16"},{"kind":"lab","title":"Alibaba Unveils Qwen4-Max (Alibaba Cloud Community)","url":"https://www.alibabacloud.com/blog/alibaba-unveils-qwen4-max-new-ai-teammate-for-work-creativity-and-discovery_603783","date":"2026-11-16"},{"kind":"lab","title":"Alibaba Cloud Model Studio: qwen4-max","url":"https://www.alibabacloud.com/help/en/model-studio/qwen4-max","date":"2026-11-16"},{"kind":"lab","title":"QwenCloud docs: Latest model, Qwen4-Max","url":"https://docs.qwencloud.com/developer-guides/getting-started/latest-model","date":"2026-11-16"},{"kind":"lab","title":"QwenCloud changelog: qwen4-max","url":"https://docs.qwencloud.com/changelog/models?model=qwen4-max","date":"2026-11-16"},{"kind":"lab","title":"QwenCloud model page: qwen4-max","url":"https://www.qwencloud.com/models/qwen4-max","date":"2026-11-16"},{"kind":"lab","title":"Qwen on X: Qwen4-Max","url":"https://x.com/Alibaba_Qwen/status/2122148127520289151","date":"2026-11-16"}],"firstBenchmark":{"date":"2026-11-17","benchmarkId":"vals-index"}},"organization":{"id":"alibaba","name":"Alibaba / Qwen","slug":"alibaba","country":null,"url":null,"color":"#FF7A00","mark":"Q"},"price":null,"answer":{"word":"#8","scope":"Best current rank: #8 in Finance, as of Nov 22, 2026.","lead":"As of Nov 22, 2026, Qwen4-Max is ranked #8 in Finance (since Nov 22, 2026; sources Finance Agent; methodology v1.0). Ranked #8 in Finance. That board does not award a crown.","confidence":"low"},"status":"BENCHMARKED","verdicts":[{"id":"v_overall_finance_qwen4-max_2026-11-22_open_RANKED","model":"qwen4-max","model_name":"Qwen4-Max","category":"finance","category_name":"Finance","slice":"overall","state":"RANKED","rank":8,"present_evidence":[],"missing_evidence":[],"coverage":null,"segments":[],"confirm":[],"margin":null,"confidence":"low","valid_from":"2026-11-22","valid_to":null,"configuration":"xhigh","rationale":"Ranked #8 in Finance. That board does not award a crown.","safeguard_flags":[],"co_sota_with":[],"challengers":[],"methodology":"1.0","label":"#8 in Finance"},{"id":"v_overall_legal_qwen4-max_2026-11-22_open_RANKED","model":"qwen4-max","model_name":"Qwen4-Max","category":"legal","category_name":"Legal","slice":"overall","state":"RANKED","rank":18,"present_evidence":[],"missing_evidence":[],"coverage":null,"segments":[],"confirm":[],"margin":null,"confidence":"low","valid_from":"2026-11-22","valid_to":null,"configuration":"xhigh","rationale":"Ranked #18 in Legal. That board does not award a crown.","safeguard_flags":[],"co_sota_with":[],"challengers":[],"methodology":"1.0","label":"#18 in Legal"}]}