Term
Qwen 3.7-Max
Qwen 3.7-Max (May 2026) is Alibabas proprietary Max flagship with a 1M-token context, positioned as the "Agent Frontier" for long, tool-intensive agent workflows.
Qwen 3.7-Max — explained in more detail
Qwen 3.7-Max is the flagship of Alibabas Qwen line, released in May 2026. An important point for context: unlike much of the Qwen family, the Max tier is proprietary — Qwen 3.7-Max is a pure API model whose weights were not published. The open, downloadable Qwen line continued separately (the open releases skipped 3.7 and later resumed with Qwen 3.8 under Apache 2.0). So equating “Qwen = open weight” is wrong for the Max tier.
Alibaba explicitly positions 3.7-Max as “The Agent Frontier”: built for long-running, autonomous workflows that execute hundreds to thousands of tool calls over hours without losing context. This fits the context window of around 1 million tokens with up to 65,536 output tokens. The API is OpenAI- and Anthropic-compatible, which eases switching from existing harnesses. In benchmarks 3.7-Max reached, among others, 80.4 on SWE-bench Verified, 60.6 on SWE-bench Pro and, at launch, around 56.6 on the Artificial Analysis Intelligence Index v4.0 — rank 5 overall and the highest-placed Chinese model there. On price the model sits at roughly 1.25 US dollars per million input and 3.75 US dollars per million output tokens.
Example / Practical context
The typical use of 3.7-Max is the sustained agentic run: an agent pursuing a complex goal over many steps — research, edit code, run tests, check results, fix up — while staying consistent for hours. Because the API is OpenAI- and Anthropic-compatible, the model can be dropped into existing agent frameworks without rebuilding the tool layer. For high-volume, tight-budget requests, Max is not the intended tool; smaller Qwen variants serve there.
Distinction from related terms
Within the Qwen line, “Max” is the strongest but closed tier — in contrast to the open Qwen models (such as Qwen 3.5 or the 3.8 open-weight line) whose weights are downloadable under a free license. That is the central difference: with Max you pay for access and performance through the API, with no self-hosting option. It differs from other providers pure frontier API models (GPT, Claude, Gemini) less in access model than in origin and its strong focus on long agentic runs. The Max tier should not be confused with the open Qwen models of the same generation.