Xiaomi MiMo undercuts frontier model pricing

This is the kind of release that makes premium AI pricing look less like a moat and more like a tax. Western labs can still charge more, but only if they can prove the extra spend buys reliability, safety, governance, and support that users actually feel.

• –Flash's 309B/15B MoE design, 256K context, and open release make the low price structurally believable.
• –The 73.4% SWE-Bench Verified result means Xiaomi is not just chasing benchmarks; it's good enough to be a serious default option for coding and agentic work.
• –Pro pushes the same pressure into the agent tier: 1M context at $1/$3 targets the long-running workflows where token bills explode.
• –Enterprise buyers may stay with Western vendors, but the negotiation baseline just moved.
• –Reddit's reaction captures the shift: people are already debating whether reliability and support are worth a 10x premium.