Qwen 3.7-Max
Alibaba's proprietary flagship, unveiled 20 May 2026 — a ~1T-parameter MoE built for hours-long agent runs, priced at roughly half of Western frontier rates.
Qwen 3.7-Max specs at a glance
| Maker | Alibaba (Qwen team) |
| Unveiled | 20 May 2026 (Apsara Summit, Hangzhou) |
| Architecture | Mixture-of-Experts · ~1T total params · ~24B active per pass |
| Context window | 1,000,000 tokens |
| API pricing | $2.50 / $7.50 per 1M input / output tokens |
| Intelligence Index | 56.6 (Artificial Analysis) |
| License | Closed-weight, proprietary |
| Best at | Long-running agent tasks with hundreds/thousands of tool calls |
Calculate API costs for this model → Compare its monthly bill against 15 other rate cards.
What's new in Qwen 3.7-Max
Qwen 3.7-Max is a strategic departure for Alibaba. The Qwen family built its reputation on open weights; 3.7-Max is proprietary and closed-weight, positioned as a paid frontier tier above the open lineup. Alibaba has announced no plan to open it.
The design brief is endurance: the model is built for tasks that run for hours and involve hundreds or thousands of tool calls without losing context — the same long-horizon agentic territory Anthropic and OpenAI are contesting, at a distinctly lower price. It scores 56.6 on the Artificial Analysis Intelligence Index.
The family moved fast around it: Qwen 3.5 (February 2026) extended to 397B parameters with 201-language support, and Qwen 3.7-Plus followed on 1 June 2026 as a multimodal agent variant at roughly one-sixth the per-token cost of Max.
Pricing and availability
At $2.50 / $7.50 per 1M tokens, Qwen 3.7-Max undercuts Claude Opus-tier rates by roughly half while claiming frontier-adjacent capability. It is live on Alibaba Cloud Model Studio, OpenRouter, Together AI and Qubrid AI, so access does not require an Alibaba Cloud account.
Who should use Qwen 3.7-Max
- Long-running agent builders — the hours-long, thousand-tool-call design target is its differentiator.
- Cost-conscious frontier users — near-frontier capability at half the Western rate card.
- Teams already in the Alibaba Cloud ecosystem — first-class Model Studio integration.
If you need open weights, look at DeepSeek V4 or Kimi K2.6 instead — 3.7-Max is the one Qwen you cannot self-host.
How Qwen 3.7-Max compares
- DeepSeek V4 — China's open-weight flagship →
- Kimi K2.6 — the open-weight coding rival →
- Claude Opus 4.8 — the Western flagship it undercuts →
- Doubao Seed 2.1 — ByteDance's competing family →
Try Qwen 3.7-Max on Model Studio
Frequently asked questions
When was Qwen 3.7-Max released?
Alibaba unveiled it at the Apsara Summit in Hangzhou on 20 May 2026, as the new proprietary top model of the Qwen family.
How much does Qwen 3.7-Max cost?
$2.50 per million input tokens and $7.50 per million output tokens — roughly half of Claude Opus-tier rates.
Is Qwen 3.7-Max open source?
No. It is closed-weight and proprietary; Alibaba has announced no plan to release it as an open model.
What architecture does it use?
Mixture-of-Experts with ~1 trillion total parameters and ~24 billion active per forward pass, and a 1M-token context window.
Where can I use it?
Alibaba Cloud Model Studio, OpenRouter, Together AI and Qubrid AI.