Qwen 3.7 Max Is Live — Alibaba's New Frontier Flagship

May 23, 2026

Qwen 3.7 Max is in the model picker today. It replaces Qwen 3.6 Max Preview as Alibaba's frontier MoE — built for long agentic loops, tool use, and deep reasoning with a 256K context window. Same price as the preview it replaces.

What you can do

• Pick Qwen 3.7 Max from the Closed Weight SOTA section in the model selector.
• Run long-horizon agentic jobs — multi-step coding, research, and tool chains where the model keeps its reasoning trace across turns.
• Use it in Swarm — Qwen 3.7 Max is swarm-eligible for parallel multi-agent runs.
• 256K context window — enough for a full codebase slice, a long brief, or a dense research corpus in one pass.
• Function calling and structured output — same agentic surface as the 3.6 Max Preview, with improved frontier quality on hard tasks.

Where this shows up

• You're running a multi-tool engineering loop and need a model that holds its train of thought across dozens of tool calls without starting over each turn.
• You want Alibaba's latest frontier tier for benchmark-style evaluation or head-to-head model comparisons — Qwen 3.7 Max is the pick.
• Your session was pinned to Qwen 3.6 Max Preview. It rolls forward to Qwen 3.7 Max on your next message — no action needed.

Try it

• "Refactor this service end-to-end — keep your reasoning visible across each step and call tools as needed."
• "Run a deep competitive analysis on these three companies. Use search, save citations, and synthesize into a report."
• "Plan and execute a multi-file bug fix. Don't lose context between tool calls."

Heads up

• Qwen 3.6 Max Preview is retired from the public picker. Existing pinned sessions autoupgrade to 3.7 Max.
• Manual pick or Swarm for now — Qwen 3.7 Max isn't in the Auto routing pool yet. Pick it explicitly, or include it in a Swarm run.
• Same pricing as 3.6 Max Preview — no credit-band change on this swap.
• Tagged "new" until June 30 — the pill in the picker is a freshness flag, nothing functional.