How do I use Qwen 3.8 in Hermes Agent?
Add a standardcompute entry under providers: in ~/.hermes/config.yaml (api https://api.stdcmpt.com/v1) with qwen/qwen3.8-max-0902 as default_model, and set model.provider to custom:standardcompute: Hermes runs on Qwen 3.8 Max with no Alibaba Cloud account, on a plan from $19 a month that stops new requests with HTTP 402 when its compute budget is spent.
# Qwen 3.8 Max by default; Flash and 27B for lighter work
# Put STANDARDCOMPUTE_API_KEY=YOUR_API_KEY in ~/.hermes/.env
providers:
standardcompute:
api: https://api.stdcmpt.com/v1
key_env: STANDARDCOMPUTE_API_KEY
transport: chat_completions
default_model: qwen/qwen3.8-max-0902
models:
qwen/qwen3.8-max-0902: { context_length: 400000, supports_vision: true }
qwen/qwen3.8-flash: { context_length: 400000, supports_vision: true }
qwen/qwen3.8-27b: { context_length: 400000, supports_vision: true }
model:
provider: custom:standardcompute
default: qwen/qwen3.8-max-0902
Use the dated qwen/qwen3.8-max-0902; the undated name is a temporary alias. hermes model, then Configure auxiliary models, moves titles and compression to Flash.
Max, Flash or 27B for Hermes
| Model ID | Fits | Per million, input / output |
|---|---|---|
| qwen/qwen3.8-max-0902 | The default agent model; reasoning from minimal to xhigh | $2 / $6 |
| qwen/qwen3.8-flash | Scheduled jobs and messaging replies that need speed more than depth | $0.15 / $0.47 |
| qwen/qwen3.8-27b | A middle option that also reads video | $0.40 / $3 |
Max and Flash at Model Studio's Singapore list price, Max with a $0.25 implicit cache read (source); 27B at OpenRouter's list price. A typical day on Max is about $4.30, a busy one $44. The Agent Burn Index measures coding agents such as OpenCode, Cline and Kilo; Hermes Agent is not broken out, so these figures use its median and 90th-percentile coding-agent day (September 13 to October 12, 2026) as a stand-in. Priced at the list rates above; your own days will differ.
Set context_length, especially for 27B
Hermes looks up a context window for each model ID. Max and Flash match its built-in entries at 1M; qwen/qwen3.8-27b falls through to a generic qwen pattern at 131,072. The config sets every model's context_length so Hermes compresses at the point you choose, not its guess.
What else to expect
- All three IDs read images, so
supports_vision: truelets Hermes send screenshots directly. - Qwen 3.8 Max always reasons: /reasoning none is raised to low rather than failing.
- Switch for one job with
/model custom:standardcompute:qwen/qwen3.8-flash.
The catch
Hermes has a built-in Qwen Cloud provider (alibaba, DASHSCOPE_API_KEY) that lists qwen3.8-max at Alibaba's own prices, plus Alibaba Coding Plan and Token Plan providers. Its picker has no Qwen 3.8 Flash or 27B, but for Max alone an Alibaba key or plan is cheaper.
Where Standard Compute fits
All three Qwen 3.8 sizes in one providers: entry, so Hermes cron jobs can run on Flash while chats stay on Max, under one monthly cap.
FAQ
My Qwen Cloud API key does not work with Hermes. What now?
Check the provider: a Model Studio key belongs in Hermes' alibaba provider (DASHSCOPE_API_KEY), a Coding Plan key in alibaba-coding-plan. A Standard Compute key only works with the standardcompute provider on this page.
Can Hermes use Qwen OAuth instead of a key?
Hermes lists a qwen-oauth provider with a browser login. It is separate from this setup, which bills a Standard Compute plan.