Alibaba Qwen launched Qwen4-Max on November 16, 2026: the flagship of the new Qwen 4 family, now generally available in the API at $1.50 input and $4.50 output per million tokens — with a 32.8% lower input price and 31% lower output price than Qwen3-Max. Its 5.6-trillion-parameter hybrid mixture-of-experts design cuts active compute to just 8% of a dense model, and it tops the LMArena WebDev leaderboard.
What changed with Qwen4-Max
Alibaba Qwen launched Qwen4-Max on November 16, 2026, as the flagship of its new Qwen 4 family. Its 5.6-trillion-parameter architecture combines a sparse mixture-of-experts design — which cuts active compute to just 8% of a dense model — with a hybrid architecture in which 1 in 8 layers is built for long-term memory, rather than attention.
The flagship is generally available in the Qwen API, and it arrives with Qwen 4's open models (Qwen4-2512, Qwen4-2512-Instruct and Qwen4-2512-Thinking) available on Hugging Face — the first open-weights model at this scale to match a proprietary flagship's output quality, according to Qwen.
The pricing is set at $1.50 input and $4.50 output per million tokens — a 32.8% lower input price and 31% lower output price than Qwen3-Max — with a 1,000,000-token context window, 2,000 tool calls per session, 512 concurrent sessions and 500,000 input tokens per minute.
In benchmarks, Qwen4-Max tops the LMArena WebDev leaderboard and scores 78 on the Artificial Analysis Intelligence Index, with leading cost-efficiency and coding scores. There is no open Qwen4-Max yet, and Alibaba Qwen has not published a date for open-weights availability.
Qwen4-Max by the numbers
- Input token price ($ per million tokens): $1.50 (Nov 16, 2026)
- Output token price ($ per million tokens): $4.50 (Nov 16, 2026)
- Intelligence Index (Artificial Analysis): 78 (Nov 16, 2026)
Why Qwen4-Max matters for business
For companies running agentic software, a flagship model that beats its predecessor on price and hits No. 1 in web coding tests directly cuts the cost of the coding, browser and tool-use workloads that dominate agent spend. Qwen 4's open models give teams a self-hosted alternative at similar quality for sensitive workloads, and the 1-million-token context window means long agent sessions and large codebases fit in one call.
Top comments (0)