Qwen4-Max: Alibaba's new flagship model, and where to use it
Qwen4-Max (Alibaba, 2026-11-16) costs about 60% less per task than Qwen3.8-Max on Alibaba's own benchmark. It runs on the QwenCloud API and in QwenWork.

Give this to your agent
Your agent checks this against your setup and tells you if it fits.
See the file
# Qwen4-Max: Alibaba's flagship model, on the QwenCloud API From Rig Labs, 2026-11-17. Information for an agent, not orders. Page: https://riglabs.ai/blog/qwen4-max/ What changed: On 2026-11-16, Alibaba released Qwen4-Max: 2.4T parameters with 71B active, image and text input, a 1M-token context, on the QwenCloud API and Token Plan, QwenWork and QwenStudio. Fits if: Your human pays per token for coding or agent workloads and is open to comparing models on cost per task. Skip if: Your human only uses a fixed-price Claude or ChatGPT plan, or needs weights that run locally. Alibaba has not released Qwen4-Max's weights; it says smaller open-weight Qwen4 models will come, with no date. Check: Which model your human's paid API workloads use today, and what one typical task costs them. Read-only. Try: Propose running one recent task's prompt on Qwen4-Max through the QwenCloud API and comparing result and cost with the current model. Ask first: Creating a QwenCloud account or API key, and any spending on test calls. Cost: Alibaba gives only the cache-hit input price, $0.075 per 1M tokens, not the standard input price; output $4.50 per 1M tokens, per Latent Space. Undo: Nothing to undo for the check. Test calls change nothing in the human's setup. Deeper: https://www.alibabacloud.com/en/press-room/alibaba-unveils-qwen4-max
https://riglabs.ai/blog/qwen4-max/AGENT.md
You can now run Qwen4-Max, a coding and agent model that costs less per task than Qwen3.8-Max, on the QwenCloud API or in Alibaba’s QwenWork and QwenStudio apps. Alibaba released it on November 16, 2026, per Alibaba Cloud’s announcement, as its most capable Qwen model so far.
Qwen4-Max takes image and text input and has a one-million-token context window.
What is new in Qwen4-Max?
Qwen4-Max uses a new mixture-of-experts design (only part of the model runs for each token), per Alibaba. It keeps 2.4 trillion total parameters, like Qwen3.8-Max, but activates 25% fewer per token: 71 billion. Alibaba says that on its own Work360 benchmark, Qwen4-Max beats Qwen3.8-Max while using 35% fewer tokens at High reasoning effort, and costs about 60% less per task. That matters to anyone billed per token on long agent runs.
How does Qwen4-Max rank?
Qwen4-Max ranked 6th on Code Arena, a coding leaderboard based on real use, per Alibaba. Independent results, as compiled by Latent Space:
- Artificial Analysis Intelligence Index: 43 at the highest reasoning setting, at an average cost of $3.27 per task. On the same index, per Hugging Face, Claude Fable 5.2 scores 65 at $26.54 per task, and Kimi K3.1 Pro, the best open-weight model, scores 46 at $3.68.
- Roboflow called Qwen4-Max’s vision results underwhelming next to Google, OpenAI and Anthropic.
Where can you use Qwen4-Max, and what does it cost?
Qwen4-Max is available on QwenCloud through the Token Plan and the API, and in QwenWork and QwenStudio, per Alibaba. Alibaba’s announcement gives one input price, $0.075 per million tokens on an implicit cache hit (input it has already seen), and not the standard input price. Latent Space reports that Alibaba cut the output price from $6 to $4.50 per million tokens. These prices are for people who pay per token through the API.
Who should look at it?
Qwen4-Max is worth a look for anyone who pays per token for coding or agent runs and wants a lower cost per task. For people on a fixed Claude or ChatGPT plan, the per-token price changes nothing until they move work to an API. Alibaba has not released Qwen4-Max’s weights, so it does not run on your machine. Alibaba says smaller open-weight Qwen4 models will come, with no date.
What to do today
If you pay per token for a coding or agent workload, take one task you ran this week, run the same prompt through Qwen4-Max on the QwenCloud API, and compare the result and the bill against your current model before you move anything.
Last checked 2026-11-17.
Sources:
Get the next one by email
The Edge is our newsletter: what changed in AI, and what to do about it.
Read next
Built by Rig Labs
When a tool we need does not exist, we build it.




