What is the Qwen3.8-Max model ID?

The model ID is qwen3.8-max on Alibaba Cloud Model Studio (DashScope), which exposes an OpenAI-compatible endpoint at https://dashscope-intl.aliyuncs.com/compatible-mode/v1 for the Singapore region and https://dashscope.aliyuncs.com/compatible-mode/v1 for Beijing.

How much does Qwen3.8-Max cost?

Launch pricing is $2.00 per million input tokens and $6.00 per million output tokens. Cached input is cheaper: $0.25 per million for implicit cache reads and $0.17 for explicit cache reads, with $2.50 per million to write an explicit cache.

Is Qwen3.8-Max available on OpenRouter?

Not at launch. OpenRouter listed Qwen3.7 Max but had no qwen3.8-max entry on August 3, 2026, so the direct DashScope endpoint is the way to reach it today. Qwen 3.7 Max remains available through OpenRouter if you prefer that routing.

Can I use Qwen3.8-Max with OpenClaw?

Yes. Add DashScope as a custom OpenAI-compatible provider with api set to openai-completions, then point the default agent model at dashscope/qwen3.8-max. On OpenClaw Launch, add your DashScope key under BYOK and enter qwen3.8-max as a custom model.

Is Qwen3.8-Max open source?

Alibaba said at launch on August 3, 2026 that it would publish the weights the following week, together with a smaller Qwen3.8-27B open-weights model. Until those land, the model is API-only.

← Home

Guide

OpenClaw + Qwen3.8-Max: Run Alibaba's 2.4T Model on Your Agent

Alibaba shipped Qwen3.8-Max on August 3, 2026 — a 2.4-trillion-parameter mixture-of-experts model with a 1M-token context window, now reachable through a plain OpenAI-compatible endpoint. Here is the model ID, the real pricing, and how to point an OpenClaw agent at it with your own key.

What Is Qwen3.8-Max?

Qwen is Alibaba's model family, and Qwen3.8-Max is its largest release to date: a sparse mixture-of-experts model at 2.4 trillion total parameters (Alibaba has not disclosed the active-parameter count). It accepts text, images, and video as input and returns text, and it carries a published 1M-token context window.

This is the release the model was pointing at when we covered the July announcement, when it was preview-only with no public model ID, no pricing, and no listing on Alibaba Cloud Model Studio. All three of those gaps closed on launch day. Alibaba also said the weights would be published the following week, alongside a smaller open-weights Qwen3.8-27B.

Qwen3.8-Max Specs and Pricing

Property	Value
Model ID	`qwen3.8-max`
Architecture	Mixture-of-experts, 2.4T total parameters
Context window	1M tokens (max input ~991K; ~983K with thinking enabled)
Max output	131K tokens (reasoning budget up to 262K)
Input modalities	Text, image, video → text out
Price	$2.00 / 1M input, $6.00 / 1M output
Cached input	$0.25 implicit read, $0.17 explicit read, $2.50 explicit write

On the benchmarks Alibaba published at launch, Qwen3.8-Max scored 86.6 on Terminal-Bench 2.1, 92.6 on GPQA Diamond, 93.0 on PaperBench, 86.1 on OSWorld-Verified, and 92.1 on OmniDocBench 1.5 — numbers Alibaba positions against frontier Western models. Treat vendor-published scores as a starting point and check independent evaluations before committing a production workload.

Where You Can Get Qwen3.8-Max Today

Alibaba Cloud Model Studio (DashScope) — the official API, and the one that matters here because it speaks the OpenAI protocol.
chat.qwen.ai — Alibaba's own chat surface, for trying the model before wiring it into anything.
QwenWork — Alibaba's workplace agent platform, which entered public beta the same day (more on that below).

Note the one that is not on the list yet: OpenRouter had no qwen3.8-max entry at launch, unlike Qwen 3.7 Max, which routes through it normally. Until it lands there, go direct to DashScope. Check the model list for what OpenClaw Launch offers as a one-click pick today.

How to Use Qwen3.8-Max with OpenClaw Launch

Qwen3.8-Max is new enough that you run it as a bring-your-own-key custom model rather than a curated one-click option. It still deploys in about 30 seconds:

Create a DashScope API key at dashscope.aliyun.com (Alibaba Cloud Model Studio).
Go to openclawlaunch.com and open the configurator.
Add your DashScope key under BYOK, and set the endpoint to Alibaba's OpenAI-compatible base URL for your region — https://dashscope-intl.aliyuncs.com/compatible-mode/v1 for Singapore, https://dashscope.aliyuncs.com/compatible-mode/v1 for Beijing.
Enter qwen3.8-max as the model.
Pick your chat platform (Telegram, Discord, WhatsApp, WeChat, or the browser gateway) and click Deploy.

Tip: A 1M-token context window is billed like any other context. If you are pasting large documents into every turn, the cached-input price is what to watch, not the headline $2 — and you can switch models any time from the dashboard or with the /model chat command, no redeploy.

Self-Hosted Configuration

If you run OpenClaw on your own server, add DashScope as a custom OpenAI-compatible provider in openclaw.json and point the default agent at it:

{
  "models": {
    "providers": {
      "dashscope": {
        "baseUrl": "https://dashscope-intl.aliyuncs.com/compatible-mode/v1",
        "apiKey": "sk-...",
        "api": "openai-completions",
        "models": [
          {
            "id": "qwen3.8-max",
            "name": "Qwen3.8-Max",
            "input": ["text", "image"],
            "contextWindow": 1000000,
            "maxTokens": 131072
          }
        ]
      }
    }
  },
  "agents": {
    "defaults": {
      "model": {
        "primary": "dashscope/qwen3.8-max"
      }
    }
  }
}

The api: "openai-completions" adapter is the right one for DashScope's compatible-mode endpoint. Use the Beijing base URL instead if your key was issued in that region — keys are not portable across regions.

Qwen3.8-Max vs Qwen 3.7 Max

	Qwen 3.7 Max	Qwen3.8-Max
Context	1M via OpenRouter	1M native
Inputs	Text	Text, image, video
Routing	OpenRouter or DashScope	DashScope direct (no OpenRouter listing at launch)
Best for	Proven, easy BYOK routing	Hardest reasoning, agentic and long-document work

If you already run Qwen 3.7 Max through OpenRouter and it is working, there is no urgency to move. Qwen3.8-Max is the upgrade to reach for when a task is genuinely bottlenecked on reasoning depth, multimodal input, or a context window you are actually filling.

What About QwenWork?

Alongside the model, Alibaba opened public beta on QwenWork, a workplace agent platform delivered through web and desktop apps and wired into DingTalk for enterprise documents, approvals, and knowledge bases. It turns a request into a deliverable — a document, a slide deck, a small deployed app.

It is a genuinely capable product, and it is also a closed loop: your agent lives inside Alibaba's suite, on Alibaba's model, reachable on Alibaba's surfaces. The tradeoff with an OpenClaw Launch agent is the opposite one — you pick the model (Qwen3.8-Max today, something else next month), you bring your own key and billing, and the agent meets your users on Telegram, Discord, WhatsApp, WeChat, or the web. Both are valid; they just answer different questions about who owns the agent.

What's Next?

OpenClaw + Qwen 3.7 — the prior generation, still the easiest Qwen to route via OpenRouter
Hermes Agent + Qwen — the same models on the Hermes framework
Compare all models — Qwen against Claude, GPT, Gemini, DeepSeek, and Kimi
Set up Telegram — put your Qwen3.8-Max agent on a chat platform