Alibaba (Qwen) FRONTIER API · Context: 1M tokens (max input 991K; 983K with thinking)

Alibaba Qwen4-Max

Alibaba's Qwen4 flagship: same 2.4T scale as Qwen3.8-Max with 25% fewer active parameters, a 1M context and lower prices ($1.50 / $4.50). Vendor-reported 46.7 on Agents' Last Exam v1.1 and 75.7 on DeepSWE v1.1, but 48.5 on Terminal-Bench 4.0, behind Opus 5.5, GPT-6.1 Sol and Kimi K3.1.

Provider
Alibaba (Qwen) (Proprietary API (no open weights for Max; smaller open-weight Qwen4 models announced as "soon"))
Architecture
Qwen4 sparse MoE (upgraded attention, residual connections and embeddings) with native image and video input; 2.4T total / 71B active
Parameters
2.4T Total / 71B Active
Context Window
1M tokens (max input 991K; 983K with thinking) (Max Output: 131,072 tokens)
API Pricing
Input: $1.50 / 1M · Cached: $0.07 / 1M · Output: $4.50 / 1M
Hardware Target
Cloud API Only

Interactive Token Cost Estimator (Qwen4-Max)

Est. API Cost (0% Cache): --/mo | With 70% Prompt Cache: --/mo
Compare all 12 models →

Architectural Strengths & Target Workloads

  • Long-horizon agent work (46.7 Agents' Last Exam v1.1 vs 43.4 for Opus 5.5, vendor-reported)
  • 35% fewer tokens and ~60% lower cost per task than Qwen3.8-Max on Work360 (vendor-reported)
  • Cached input at $0.075 / 1M for long sessions

Primary Engineering Workloads: Planner / Long-Running Research Agents · Agentic Knowledge Work · Multimodal Document & Video Understanding.

Frequently Asked Questions (Qwen4-Max)

How much does Alibaba Qwen4-Max cost per 1M tokens?

Alibaba Qwen4-Max is priced at $1.50 per 1M uncached input tokens, $0.07 per 1M cached input tokens, and $4.50 per 1M output tokens (QwenCloud / Alibaba Cloud Model Studio (model id qwen4-max), international list price; OpenAI- and Anthropic-compatible APIs. Explicit cache writes $1.875 (5 min) / $3.00 (1 h) per 1M. Output with thinking up to 262K.).

What is the context window of Alibaba Qwen4-Max?

It supports up to 1M tokens (max input 991K; 983K with thinking) (1,000,000 tokens) and a maximum generation output of 131,072 tokens.

Full TechPulse Engineering Analysis

Qwen4-Max: Alibaba's 2.4T-Parameter MoE Ships With a 1M Context at $1.50 / $4.50 per Million Tokens

Alibaba's new flagship keeps the 2.4T scale of Qwen3.8-Max but activates 25% fewer parameters per token, and it costs less. Its strongest results are on long-horizon agent benchmarks. On terminal coding it still trails Opus 5.5, GPT-6.1 Sol and Kimi K3.1.

Read Full Technical Dispatch (5 min read) →
Risograph-style illustration of a long colonnade of pale cream columns standing on a very long plinth, with only three of the columns painted in ochre