QWEN3-MAX

Qwen3-Max: trillion-parameter flagship with thinking mode

Alibaba's Qwen3 flagship. Trillion parameters. Thinking mode for deeper reasoning. 262K context. Strong agentic programming and tool calling. $1.20/$6 per million tokens.

Try Qwen3-Max on Council See it in a council

Qwen3-Max is the trillion-parameter flagship of the Qwen3 family — Alibaba's prior top-of-stack model before Qwen3.6-Max shipped in April 2026. It exposes a thinking mode that produces a longer internal chain-of-thought before answering, analogous to GPT-5.5's reasoning-high or Claude Opus's extended thinking. The 262K-token context window handles large codebases and long documents without RAG. Native web search via DashScope is on. Priced at $1.20 per million input tokens and $6 per million output, it's modestly more expensive on input than the newer Qwen3.6-Max ($1.04) but cheaper on output ($6 vs $6.24). The case for: agentic programming, tool-heavy workflows, and reasoning tasks where you want thinking-mode depth at a fraction of GPT-5.5's price. The case against: Qwen3.6-Max now leads on six coding benchmarks at similar pricing — for greenfield work pick the newer model. Keep Qwen3-Max in the rotation when you specifically want its thinking-mode behavior.

Specs

ProviderAlibaba (Qwen)
Parameters~1 trillion
Context window262,144 tokens
ReasoningThinking mode
Web searchNative (DashScope enable_search)
Price (in / out per 1M)$1.20 / $6.00
MultimodalText only

Best at

Where it loses

Frequently asked questions

Should I use Qwen3-Max or Qwen3.6-Max?

For greenfield coding work, pick Qwen3.6-Max — it leads six top coding benchmarks at similar pricing. Pick Qwen3-Max when you specifically want thinking-mode behavior or already have prompts tuned to it.

What is thinking mode?

A reasoning mode where Qwen3-Max produces a longer internal chain-of-thought before responding. Better on hard math, logic, and architecture problems at the cost of latency.

How big is Qwen3-Max?

Approximately one trillion parameters — Alibaba's largest publicly announced model in the Qwen3 family.

How much does Qwen3-Max cost?

$1.20 per million input tokens and $6.00 per million output. Council AI bundles it inside monthly plan budgets.

Does Qwen3-Max support function calling?

Yes — tool calling and agentic workflows are a core strength.

Is Qwen3-Max open source?

No — the Max tier is closed weights. Smaller Qwen3 models in the family are open-source.