Big Model Value Wars - DeepSeek V4 Pro vs MiMo-V2.5-Pro vs MiniMax M3

For those who sometimes boost their local model use with openrouter options, or the madlads who have the infrastructure to actually run those locally, it feels like those three model have the edge in best bang for your buck. How then do you decide which one to use? Do you have a strong opinion on which model is best? Or do you have specific use cases? Personally I'm thinking for agentic and coding use cases, paired with Hermes Agent (now trying Desktop) as well as both Qwen 3.6 27b and 35b. Which model do you recommend of the three and why? Or do you have preferences outside those three?