MiMo-V2.6-Pro-UltraSpeed

FoneArena's coverage of the MiMo-V2.6 launch confirms that Xiaomi released and open-sourced the series and explicitly names MiMo-V2.6-Pro-UltraSpeed as part of the rollout, alongside the natively omnimodal MiMo-V2.6-Pro and MiMo-V2.6-Flash. The article reports MiMo-V2.6-Pro scored 46.32 on the Artificial Analysis Intel Additional detail includes training cost figures (about $0.85M for Flash and $2.62M for Pro), average pass-rate improvements (25% and 12% in relative terms), and DeepSWE v1.1 gains (Flash from 48.8 to 65.68; Pro from 58.4 to 72.57). Xiaomi scaled RL compute via larger batches (1,568 samples per update, up to 1M-token c

Xiaomi's official MiMo-V2.6 launch page, dated September 22, 2026, introduces the MiMo-V2.6 series as natively omnimodal models, with MiMo-V2.6-Pro-UltraSpeed explicitly named as a rolling-out variant that delivers "up to 20x faster output speed at the same quality" for users requiring extreme generation speed. The pag The launch page frames the series around recursive self-improvement via reinforcement learning on verifiable tasks, with MiMo-V2.6-Pro-UltraSpeed positioned as a latency-optimized variant for high-throughput deployment scenarios. Xiaomi presents extensive benchmark results (DeepSWE v1.1, ProgramBench, MiMo Code Bench,

Flowtivity's September 22, 2026 analysis confirms Xiaomi released MiMo-V2.6-Pro-UltraSpeed alongside MiMo-V2.6-Pro and MiMo-V2.6-Flash, explicitly naming UltraSpeed as "a latency-optimised build of the same checkpoint" with the goal of much faster output at comparable quality. The piece reports MiMo-V2.6-Pro at 46 on t Beyond UltraSpeed confirmation, the article cites Xiaomi's launch claim that at comparable intelligence the Pro model costs roughly one-twentieth to one-sixtieth of leading international models, and frames MiMo-V2.6-Pro as trading wins with much pricier frontier models on agentic workloads. It flags the MIT license and