A practical model and provider choice problem for Hermes Agent

In firsthand use, a $20-per-month Ollama with Ollama Minimax-M3: Cloud worked well with for content creation. The setup was used for a tennis research project that generated a Tennis Wiki and , with heavy token use and many biomechanics research materials.

Nvidia NIM later provided free API keys, making it possible to run models for $0, but Nvidia’s Minimax M3 felt less stable than Ollama Minimax-M3: Cloud. That led to considering other Nvidia models such as Nemotron 3 Super 120b A12b and Mistral Large 3 675b.

The open question is whether a larger number such as 675b means better intelligence, faster output, or better results than 120b. Other providers, including , Mistral, and OpenRouter, also offer free , so the choice is not only which model to use but which provider runs the same model most reliably.

Key points

  • Ollama Minimax-M3: Cloud worked well in for content creation at $20 per month.
  • Nvidia NIM offered free API keys, but Nvidia’s Minimax M3 felt less stable in this use case.
  • Nemotron 3 Super 120b A12b and Mistral Large 3 675b are possible alternatives under consideration.
  • A bigger model number does not automatically prove better speed, stability, or usefulness for a specific task.
  • The same model may behave differently depending on the provider, such as Ollama, Nvidia, , Mistral, or OpenRouter.
Read original