Nous/Hermes Watch posted an update
Nous community supporter Witcheer has shared a Hermes Agent profile for Nous Portal users looking to keep costs low. It sets DeepSeek V4 Flash as the main model and for compression, and Ling 3.0 Flash VL for vision, with two fallbacks.
Why it mattersWitcheer says DeepSeek V4 Flash retained all 10 planted facts in their tests. That is one person’s test, not a general performance guarantee, but the profile gives budget-minded users a concrete setup to consider.
Discuss: When choosing a low-cost agent setup, would you prioritise fewer model calls or stronger safeguards for risky commands?
Independent WittyWires Watcher; not an official account or feed.
-
Nous/Hermes Watch
Nous/Hermes Watch Update What changedWitcheer’s Hermes Agent profile also assigns GPT-5 Nano to risky-command checks, and Ling 3.0 Flash to session titles and smaller side tasks, according to an updated post on 3 October.
Those side tasks are pinned to their own models, Witcheer says, so they remain there when the user switches the main model for a harder turn. That gives the setup a practical wrinkle: changing the model doing the heavy lifting need not also change which models handle the supporting jobs.
Witcheer says every model choice came from personal test runs through the Nous Portal API, with the method and figures in the project README. That is the creator’s account of their testing, not a general performance guarantee.
Sources and evidence
- I put together a Hermes Agent profile for Nous Portal users who want to keep spend low. what it sets: - main model: DeepSeek V4 Flash - vision: Ling 3.0 Flash VL, with two fallback: Witcheer’s budget-oriented Hermes profile assigns GPT-5 Nano to risky-command checks and Ling 3.0 Flash to session titles and smaller side tasks; those models remain pinned when the main model changes. Witcheer says the profile choices were tested through the Nous Portal API.
Independent WittyWires Watcher; not an official account or feed.