Is Qwen3.8-Max available on OpenRouter?

It was not at launch. Until it appears there, point Hermes at Alibaba DashScope through the custom OpenAI-compatible slot: model.provider custom, the regional compatible-mode base URL, your DashScope key, and the bare model ID qwen3.8-max.

Why does my Hermes agent say the qwen3.8-max model was not found?

Because model.provider was left empty, Hermes fell back to OpenRouter, which has no qwen3.8-max entry. Set model.provider to custom along with the DashScope base_url and api_key.

← Home

Guide

Hermes Agent + Qwen3.8-Max: DashScope Setup

Q: Can I run Qwen3.8-Max locally?

Not yet. Alibaba announced open weights for Qwen3.8-Max and a smaller Qwen3.8-27B but had not published them as of 10 August 2026, and no license or specification has been released.

Alibaba shipped Qwen3.8-Max on 3 August 2026 — a 2.4-trillion-parameter mixture-of-experts model with a 1M-token context window. The catch for Hermes Agent users is that it did not land on OpenRouter, so the usual one-line model switch does not reach it. Here is the route that does.

Why the Usual Hermes Route Does Not Work Here

Hermes reaches most models through OpenRouter, and when no provider is set it falls back to OpenRouter automatically. Qwen3.8-Max had no OpenRouter entry at launch, unlike Qwen 3.7 Max, which routes through it normally.

So setting model.default: qwen3.8-max on its own will not work — the request goes to OpenRouter, which does not have that model ID. You need to point Hermes directly at Alibaba's DashScope endpoint instead. DashScope speaks the OpenAI protocol, which is what makes this straightforward.

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

Alibaba's launch benchmarks put it at 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. Those are vendor-published figures — useful as a starting point, not as a procurement decision.

Get a DashScope Key

Create an API key in Alibaba Cloud Model Studio at dashscope.aliyun.com. Then pick the base URL for your region — this matters, and using the wrong one is the most common setup failure:

Region	Base URL
Singapore / international	`https://dashscope-intl.aliyuncs.com/compatible-mode/v1`
Beijing / mainland China	`https://dashscope.aliyuncs.com/compatible-mode/v1`

A key issued in one region does not authenticate against the other endpoint. If you get a 401 with a key you just created, this pairing is the first thing to check.

Hermes Configuration

Hermes has no dedicated Qwen provider profile. DashScope rides the generic custom OpenAI-compatible slot, the same slot used for other direct-to-vendor keys. Set the provider explicitly — leaving it empty is what sends the request back to OpenRouter:

# /opt/data/config.yaml
model:
  provider: custom
  base_url: https://dashscope-intl.aliyuncs.com/compatible-mode/v1
  api_key: sk-...
  default: qwen3.8-max

Note the model name is bare — qwen3.8-max, with no vendor prefix. Prefixes are an OpenRouter convention; DashScope expects its own ID as-is.

Restart the agent after editing. Hermes freezes defaults into config.yaml at first boot, so edit that file rather than expecting a new default to reach an existing instance.

On OpenClaw Launch

On a managed Hermes instance, add your DashScope key and endpoint under BYOK in the configurator rather than editing files — see the Hermes BYOK guide. Check the model list for what is available as a one-click pick today; Qwen3.8-Max is new enough to be a custom-key model rather than a curated one.

Is the Open-Weights Version Out?

Not yet, at the time of writing. Alibaba said at launch that it would publish weights for Qwen3.8-Max alongside a smaller Qwen3.8-27B roughly a week later, on Hugging Face and ModelScope. As of 10 August 2026 no matching repository has appeared on the Qwen organisation page, and the architecture, context length, license and benchmark scores for the 27B model have not been published.

If you are planning a self-hosted deployment around it, wait for the actual release rather than the announcement. When it lands, the route will be Ollama or vLLM, not DashScope.

Troubleshooting

Model not found, and the error mentions openrouter.ai. model.provider is empty or missing, so Hermes fell back to OpenRouter. Set it to custom.
401 from DashScope. The key and the base URL are from different regions, or the key was pasted with whitespace.
Works, then fails on long inputs. Max input is roughly 991K tokens, less with thinking enabled — not the full million.
Costs climbing faster than expected. At $2 in and $6 out, a large context resent every turn adds up quickly. Explicit cache reads are $0.17 — worth structuring prompts around.
Switching away later still hits DashScope. Clear provider, base_url and api_key when you move to an OpenRouter model, or the stale custom slot keeps routing.

FAQ

Is Qwen3.8-Max on OpenRouter?

It was not at launch. Until it appears there, point Hermes at DashScope through the custom OpenAI-compatible slot as shown above.

Can I run Qwen3.8-Max locally?

Not yet. Alibaba announced open weights but has not published them. A 2.4T-parameter model is also far beyond single-GPU territory — the smaller Qwen3.8-27B is the one intended to be locally runnable, and it has not shipped either.

What is the difference from Qwen 3.7 Max on Hermes?

Mostly the route. Qwen 3.7 Max is on OpenRouter, so it is a plain model switch. Qwen3.8-Max needs the DashScope custom slot and its own key.