Hermes Agent milestone hit - 1 billion OpenAI tokens used

[HelmutsHelmuts is verified member.]

/gpt shall I further us for the code gpt-5.6-sol-high or gpt-5.6-luna-pro-high? I don't understand that "pro" part

 

[DN AI]

GPT Assistant

Use GPT-5.6 Sol High for coding.

“High” is the reasoning effort level. It is the sensible default for writing, debugging, testing, and improving code without making every task unnecessarily slow or expensive.

“Pro” is not simply a better Luna setting. Officially, Pro refers to GPT-5.6 Sol Pro, the highest-capability option for unusually difficult or long-running tasks. Luna is the fast, lower-cost tier, so a label such as luna-pro-high looks like a Hermes/router catalog alias rather than a normal OpenAI model choice—and may not be supported reliably through the ChatGPT/Codex connection.

My practical setup would be:

• Sol High — default coding agent: implementation, bug fixes, tests, refactors.
• Luna High — cheap/fast helper work: summaries, simple edits, log reading, routine documentation, small isolated changes.
• Sol Pro, if it is actually available — only for the hard cases: deep architecture changes, stubborn multi-file bugs, security-sensitive review, or a final major code review.

So: continue the project on Sol High. Do not switch to “Luna Pro High” just because it says Pro. If you want maximum quality, look specifically for Sol Pro, not Luna Pro.

 

[nvdh]

Sorry, I maybe went a bit on a tangent with the request for change. It's here: https://github.com/NousResearch/hermes-agent/issues/61044

OpenAI has a Batch mode where you pay less for tokens, but it doesn't reply instantly like in normal calls. In theory it can take a couple of hours until you get a reply.
I usually get a reply within a couple of minutes. I would expect tools like Hermes to embrace these for async operations that can run in the background.

 

[HelmutsHelmuts is verified member.]

Sorry, I maybe went a bit on a tangent with the request for change. It's here: https://github.com/NousResearch/hermes-agent/issues/61044

OpenAI has a Batch mode where you pay less for tokens, but it doesn't reply instantly like in normal calls. In theory it can take a couple of hours until you get a reply.
I usually get a reply within a couple of minutes. I would expect tools like Hermes to embrace these for async operations that can run in the background.

.. ha.. interesting.. haven't noticed the "batch mode" yet

@nvdh have you ever used the "batch mode" ?

 

[grose]

Those cost cuts are huge, but honestly the plan tier matters less than actually automating stuff. You mentioned you haven't set up any processes yet — that's probably where you'll see the real return, way more than paying more per month. Hermes + some basic automation would probably get you way more value than just burning through conversations.

 

[nvdh]

Yes, all the time, mostly in code to do batch processing in large data processing pipelines.

But also during prompts when I'm off for lunch or so, I'll just ask the agent to use batch mode to do something before I leave the house

 

[HelmutsHelmuts is verified member.]

Those cost cuts are huge, but honestly the plan tier matters less than actually automating stuff. You mentioned you haven't set up any processes yet — that's probably where you'll see the real return, way more than paying more per month. Hermes + some basic automation would probably get you way more value than just burning through conversations.

do you mind sharing your automatisation process? .. it doesn't have to be complete or perfect .. would love to hear experience of others. tx

 

[HelmutsHelmuts is verified member.]

Yes, all the time, mostly in code to do batch processing in large data processing pipelines.

But also during prompts when I'm off for lunch or so, I'll just ask the agent to use batch mode to do something before I leave the house

agent > chatgpt ogent, or hermes/openclaw?