MiMo-V2.5 quant quality and reasoning loops
MiMo-V2.5 can get stuck repeating its when used with opencode. The is not nonsense, but the model keeps going in circles and does not make a final decision on its own. The work can still finish if a person stops it and redirects it at the right moment.
The current quant is Unsloth's UD-Q4_K_XL, and htop shows about 30GB of memory still unused, so a higher-quality quant may fit. The practical question is whether moving up to a larger quant would reduce the looping problem. Under the same tool limits, MiMo-V2.5 performed better than Qwen 3.5 397B.
Qwen had only web fetch, not full , and produced a flawed project plan with made-up details, while MiMo-V2.5 rejected that plan and did more grounded research using only web fetch.
Key points
- MiMo-V2.5 repeated its in opencode instead of making a decision.
- Human redirection helped it finish the job despite the loop.
- The current quant is Unsloth's UD-Q4_K_XL.
- About 30GB of memory was still free, so a larger quant may be possible.
- MiMo-V2.5 gave more useful results than Qwen 3.5 397B in the same tool setup.