AlibabaAlibaba·💬 Text Generation·New

Qwen 4 Max

ReasoningVisionCodeFunction CallingWeb Searchanonymized
🧠 Try in Intelligence →Try on Venice.ai ↗
Quick reference
Qwen 4 Max — TLDR
  • 📏 1M-token context window, up to 393,216 output tokens
  • 🧠 Thinking on by default; reasoning effort low, medium or xhigh
  • 👁️ Accepts text, images and video; returns text
  • 🔧 Function calling, web search, structured JSON outputs supported
  • 🏢 Alibaba's next-generation flagship in the Qwen Max line
  • 🆕 Output budget now counts thinking plus the answer together
  • 🌐 OpenAI and Anthropic API compatible endpoints
  • 🎯 Targeted at coding, office work, front-end design, agents
💰 Pricing
$1.88 / $5.63
per 1M · input / output
📏 Context
1M tokens
📅 On Venice since
Nov 16, 2026
0 days ago
Provider

Alibaba Group is a Chinese multinational technology company founded in 1999 and headquartered in Hangzhou, Zhejiang. Originally built around e-commerce and cloud computing, Alibaba has become one of the most prolific contributors to open-weight AI research,…

Read full profile →
78 models on Venice
33 video · 24 text · 9 image · 8 inpaint · 2 embedding · 2 tts
Since Jan 11, 2025

About this model

Qwen 4 Max is the flagship tier of Alibaba's Qwen 4 generation, first exposed publicly as a preview model identified as qwen4-max-preview. Alibaba Cloud Model Studio describes it as "the next-generation flagship multimodal model in the Qwen family, combining text generation, image and video understanding, and deep reasoning," designed for coding, office productivity, frontend design and complex agentic tasks. It takes text, images and video as input and returns text, and carries a one-million-token context window.

Compared with its same-family predecessor Qwen 3.8 Max, the documented headline capacity is a maximum output of 393,216 tokens alongside the 1M-token context and text, image and video input. The accounting also changed — starting with this model, the max tokens parameter limits thinking and the final answer combined, where earlier Qwen models (including Qwen 3.7 Max) applied the cap to the answer alone. Ports from older code may therefore need a larger budget.

Reasoning is enabled by default, with an effort control accepting low, medium or xhigh (xhigh being the default); Alibaba's docs warn against combining that parameter with an explicit thinking budget. Tooling support covers function calling, web search and other built-in tools, plus structured outputs in JSON Object and JSON Schema modes, served through endpoints compatible with the OpenAI and Anthropic formats.

No parameter count, weights release, or benchmark results are listed on the official model page, so this page avoids performance figures. For multimodal work on open weights, the earlier Qwen3 VL 235B remains the vision-language reference point in the family.

This About section is AI-generated from public sources (Claude Opus 5), with no human editing. It may contain inaccuracies — verify critical details against the sources listed above.

Data sources: Venice API · HuggingFace · Wikipedia — enrichment updated 12h ago