Skip to content
vavio

Can my scheduled Hermes job use a cheaper model than my chat?

Yes. Hermes Agent can use one model for chat and another for scheduled jobs: set a model on an individual job, or set cron.model as the default for unpinned jobs. If you set neither, the job uses your main model when it fires, so changing your chat model can change the model used by your schedules too. On Vavio, model usage remains separate from the computer’s hosting bill.

This answer follows the official Hermes v0.23.1 cron guide, commit 3b42157, checked on 15 November 2026. It explains the documented controls; we did not run a scheduled-job comparison or measure savings. A cheaper model is useful only if it still completes your job correctly. Check your provider’s current rates and the actual output before moving important recurring work.

Which model will my scheduled job use?

Hermes resolves the model when the job fires, in this order:

Setting Jobs it affects What happens when you change your chat model?
A per-job model pin That job It keeps its pinned model.
cron.model and cron.model_provider Unpinned jobs They keep the cron default.
Main model, selected with hermes model or /model Jobs with neither setting above They follow the main model on their next run.

For example, you might want an inexpensive model for a daily summary and a stronger model for interactive work. A cron default separates those choices without changing your chat model. A per-job pin makes sense when just one schedule needs a different route.

Each cron run starts a fresh agent session. The next day’s summary does not inherit yesterday’s chat history. That is a different cost question from why a long Hermes conversation grows more expensive.

How do I change one job’s model?

The owner can choose a specific model and provider in the Hermes dashboard or with the Hermes CLI. In a shell where your Hermes profile is available, list your jobs and edit the intended one:

hermes cron list
hermes cron edit <job_id> --provider <provider> --model <model>

Replace the angle-bracket placeholders with the job ID and a provider/model you have configured. Its credentials must be available for unattended use. This is an upstream Hermes command, not a claim that a Vavio page exposes every cron setting.

hermes cron edit <job_id> --pin does something narrower: it locks that job to the current main model. It does not select a cheaper model for you. --unpin removes the job’s pin, after which it follows the cron default if configured, otherwise the main model.

Asking Hermes to pin the current main model is supported by its cronjob_manage tool when you explicitly request it. Choosing a different inference model is owner-controlled: that tool cannot point the job at another model. Use the dashboard or CLI for that choice.

How do I set a default for all unpinned jobs?

Set the cron model and its provider in the active profile’s configuration:

cron:
  model: "your-model-id"
  model_provider: "your-provider"

Use your configured route’s actual values. The guide also documents hermes config set cron.model <model> for setting the model. Jobs with their own pins keep those pins; the default applies to jobs without one. Switching models in chat does not replace this cron default.

There is a reliability tradeoff between a default and a pin. An unpinned job can use the configured global fallback-provider chain if its resolved route fails. A job with its own model, provider or base URL does not use that alternate-provider chain: if its pinned route fails, the run fails. Both kinds of job may rotate to another credential for the same provider when a credential pool is configured.

If your reason for a pin is keeping spending predictable, inspect any fallback chain too. An unpinned job may run on a different provider and model after a failure. A pin controls the route; it does not guarantee a successful result or a fixed bill.

Can I lower reasoning effort for just this job?

Yes. A job can carry its own reasoning-effort setting independently of its model pin:

hermes cron edit <job_id> --reasoning-effort low

That job’s setting overrides both the global agent.reasoning_effort and per-model agent.reasoning_overrides. Pass an empty string on edit to clear it and follow the configuration again:

hermes cron edit <job_id> --reasoning-effort ""

The documented levels are none, minimal, low, medium, high, xhigh, max and ultra; none disables thinking. Providers clamp or omit levels their model does not support, so a label is not a promise of an exact reasoning budget. The setting has no effect on jobs that skip the model entirely. It is not exposed through the agent’s cron-management tool.

Try a lower effort on a bounded job with output you can check. Compare whether it found the right information, followed the required format and finished the task. We have not measured how this changes cost or quality for any model.

Do toolsets affect the job’s model input?

Yes. Enabled tools contribute their definitions to the model’s prompt. The cron guide specifically warns that carrying browser and delegation tools into every small fetch-and-summary job adds tool-schema input on every model call.

A per-job enabled_toolsets list takes precedence over the cron-platform selection in hermes tools; that selection takes precedence over built-in defaults. For a job that only needs web and files, the guide shows:

enabled_toolsets=["web", "file"]

This is a field accepted by the cron create/update tool, not a standalone shell command. Keep the tools the job actually needs. Removing a required tool can leave you with a smaller prompt and an unfinished task. If the cron-platform configuration cannot be read, the documented behavior is a recorded run error, rather than silently enabling every tool.

What should I change first?

Choose one non-critical job whose successful result you can recognize. Record its current route, then choose either a per-job pin or a cron default depending on whether the change should affect one job or the unpinned fleet. Check the result and your provider’s usage record before extending the change. Change reasoning effort and toolsets separately so you can see which choice caused a difference.

This article explains where the controls are; it does not rank models. Using Hermes without an OpenAI or Anthropic account explains provider routes, while the monthly cost guide separates hosting, model usage and any assistant subscription you already have.

Vavio supplies a managed remote computer for Hermes at $14.99 per month. You keep control of Hermes’s model and configuration, with model usage billed separately. If you want the computer managed while your scheduled work runs away from your laptop, compare the current hosting offer. If you already run a server comfortably, the same upstream Hermes settings apply to your own installation.

Sources

Written by Vavio with AI assistance and checked against pinned official Hermes v0.23.1 documentation. This is a documentation review, not a measured savings or scheduled-job trial.

More from the Vavio blog: Hermes Agent guides