MiMo-V2.6-Pro-UltraSpeed

Flowtivity's analysis explicitly names MiMo-V2.6-Pro-UltraSpeed as a latency-optimised build of the same Pro checkpoint, released on September 22, 2026. MiMo-V2.6-Pro is a 1.02 trillion-parameter sparse mixture-of-experts with 42B active per token, while MiMo-V2.6-Flash offers 309B total parameters with 15B active per token for efficiency-focused workloads. All checkpoints ship under an MIT license with a 1M-token context window. MiMo-V2.6-Pro reaches a 46 score on the Artificial Analysis Intelligence Index at roughly US$0.13 per task, placing it on the intelligence-versus-cost Pareto frontier and one point behind GPT-5.6 Sol. Xiaomi positions Pro as trading wins with leading closed models on agentic workloads like tool use, terminal work, and computer use at a fraction of the inference cost. The UltraSpeed variant layers latency optimisation onto the Pro checkpoint for speed-sensitive use cases.

FoneArena's coverage of the MiMo-V2.6 launch explicitly names MiMo-V2.6-Pro-UltraSpeed as part of Xiaomi's rollout. The release focuses on scaling reinforcement learning compute on verifiable, complex tasks using a fully asynchronous architecture that supports up to 1 million-token context and processes 3.5–3.7 billion tokens per step across 1,568 samples per update. Xiaomi streamed the production RL run live over roughly six days. The article reports DeepSWE v1.1 gains from 48.8 to 65.68 for Flash and 58.4 to 72.57 for Pro, with average pass rates rising 25% and 12% in relative terms. Xiaomi also details defenses against reward hacking, including reward design, adversarial evaluation, anomaly detection, and cross-checking between verifiers, and is open-sourcing the supporting technical components. The series retains the V2.5 API pricing structure.

Xiaomi officially rolled out the MiMo-V2.6-Pro-UltraSpeed variant alongside its open-source MiMo-V2.6 series on September 22, 2026. Xiaomi describes UltraSpeed as delivering up to 20x faster output speed at the same quality as MiMo-V2.6-Pro, targeting users who need extreme generation speed. The broader V2.6 release also included natively omnimodal MiMo-V2.6-Pro and MiMo-V2.6-Flash models. MiMo-V2.6-Pro scored 46.32 on the Artificial Analysis Intelligence Index and posted competitive results across DeepSWE v1.1, ProgramBench, MiMo Code Bench, Toolathlon, Terminal Bench 4.0, JobBench, and the CyberGym suite. The Pro and Flash checkpoints both completed 30 RL steps over ~750,000 trajectories in under six days, with reported training costs of about $0.85M for Flash and $2.62M for Pro. Xiaomi retained V2.5-series API pricing for the new release.