Hermes Field Guide

The four chapters of the old guide that said what the official docs already say were deleted. This is what is left: the part that took real work, and that the docs deliberately do not do, because a vendor cannot tell you which of its own models to buy.

Why the auxiliary slots are the whole question

Hermes has a main model for the visible agent loop, and separate auxiliary slots for the side jobs: context compression, vision, web extraction, approvals, session titles. Every auxiliary slot defaults to the main model unless you override it.

That default is convenient, and it is how people accidentally spend real money having a frontier model write conversation titles. Compression in particular runs constantly and on long inputs. Setting one cheap auxiliary model is the single highest-leverage line in config.yaml.

model:
  provider: 'nous'
  default: 'anthropic/claude-sonnet-4.6'
auxiliary:
  compression:
    model: 'google/gemini-3.1-flash-lite'
  web_extract:
    model: 'google/gemini-3.1-flash-lite'
  vision:
    model: 'google/gemini-3.1-flash-lite'

Prefer hermes model over hand-editing that file while you are learning: the wizard also sets the context length and the auth mode, and both are easy to get silently wrong.

Four starting combinations

Profile	Main model	Auxiliary / compression	Use it when
Low-cost learner	DeepSeek: DeepSeek V4 Flash 0423 `deepseek/deepseek-v4-flash` $0.05 / $0.14	DeepSeek: DeepSeek V4 Flash 0423 `deepseek/deepseek-v4-flash` $0.05 / $0.14	You are learning the loop, the tools and the prompts. Burn tokens, not money.
Balanced daily	Claude Sonnet 4.6 `anthropic/claude-sonnet-4.6` $3.00 / $15.00	Google: Gemini 3.1 Flash Lite `google/gemini-3.1-flash-lite` $0.25 / $1.50	You want reliable tool use without paying the main model to write titles.
Quality first	Claude Opus 4.6 `anthropic/claude-opus-4.6` $5.00 / $25.00	Google: Gemini 3.1 Flash Lite `google/gemini-3.1-flash-lite` $0.25 / $1.50	The task is hard and failure costs more than tokens do.
Long-context work	Google: Gemini 3.1 Pro Preview `google/gemini-3.1-pro-preview` $2.00 / $12.00	Google: Gemini 3.1 Flash Lite `google/gemini-3.1-flash-lite` $0.25 / $1.50	Large documents, or sessions that run all day.

USD per 1M input / output tokens, read from the live catalogue on 2026-09-20. Checked against Hermes Agent v0.21.3 (v2026.9.14), 402 models listed.

These are starting points, not rankings. Do not read "cheap auxiliary" as "always good enough" — compression and web extraction can drop details that mattered, and you will not be told. Test a proposed auxiliary model against your own workload before you trust it with a long session.

Where these numbers come from

Every price above is read from the Nous Portal catalogue and written into upstream.json. scripts/freshness.mjs re-reads the catalogue and fails the build if any of them has moved, so this table is either correct or this site does not deploy.

That check exists because of this exact page. The version of it that shipped in ponzgpt/hermes-launch recorded a promotional discount, the promotion ended, and five of its six prices quietly became 80% of the real ones for weeks. Nobody noticed, because nothing was watching. Now something is.

The catalogue is the source of truth for what a model costs; how the slots behave is upstream's: configuring models. Portal plans bundle credits, so what you actually pay may be less than the per-token rate. None of this is financial advice and none of it is a guarantee.

Pinned ids, never `-latest`

The catalogue publishes ~vendor/model-latest aliases alongside pinned ids. This guide names pinned ids only. An alias keeps its name while its price and its behaviour move underneath you, which is the precise failure this page is trying not to repeat.