MiMo V2 Omni Looks Less Censored

My take: for developer-facing models, “less censored” often means “more usable,” but only if the model still stays truthful and steerable.

• –The Omni response is clearly more practical in the example because it answers the user instead of shutting the conversation down.
• –That kind of behavior tends to help in coding and agent workflows, where dead-end refusals are costly.
• –The tradeoff is that softer guardrails can also mean more risk, more policy violations, and less consistent behavior across prompts.
• –This is a single anecdote, so I would not overfit on it; it’s a signal about tuning, not a full capability verdict.
• –If MiMo V2 Omni is meant to be the more multimodal agent model, it may simply need better calibrated safety behavior rather than stronger blanket censorship.