kind,metric,domain,unit,pro,flash,source_url,access_date,evidence,note
training,RL phase reported cost,spend,USD millions,2.62,0.85,https://mimo.mi.com/docs/en-US/news/latest/v2-6,2026-09-29,Xiaomi estimate,Reported RL phase only; excludes earlier and later development.
training,RL optimizer steps,volume,steps,30,30,https://mimo.mi.com/docs/en-US/news/latest/v2-6,2026-09-29,Xiaomi report,Both runs said to complete 30 steps.
training,Prompts per step,volume,prompts,1568,1568,https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL,2026-09-29,Xiaomi model cards,"Both cards state 1,568 prompts per step."
training,Rollouts per prompt,volume,rollouts,16,16,https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Flash-RL,2026-09-29,Xiaomi model cards,Both cards state 16 rollouts per prompt.
training,Nominal rollout slots,volume,rollouts,752640,752640,https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL,2026-09-29,Arithmetic from Xiaomi configuration,"30 × 1,568 × 16; nominal slots, not independently observed completed trajectories. Release wording of ~750,000 cumulative trajectories is ambiguous."
benchmark,DeepSWE v1.1,Code Agent,reported score points,71.9,67.9,https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL#3-evaluation-results,2026-09-29,Xiaomi model-card comparison,Same table and named benchmark/version; evaluation harness details are not fully specified.
benchmark,ProgramBench,Code Agent,reported score points,26.5,26.0,https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL#3-evaluation-results,2026-09-29,Xiaomi model-card comparison,Same table and named benchmark/version; evaluation harness details are not fully specified.
benchmark,AutomationBench v1.0.6,General Agent,reported score points,53.1,52.3,https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL#3-evaluation-results,2026-09-29,Xiaomi model-card comparison,Same table and named benchmark/version; evaluation harness details are not fully specified.
benchmark,Terminal Bench 4.0,General Agent,reported score points,34.9,28.8,https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL#3-evaluation-results,2026-09-29,Xiaomi model-card comparison,Same table and named benchmark/version; evaluation harness details are not fully specified.
benchmark,Terminal Bench 2.1,General Agent,reported score points,89.9,87.6,https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL#3-evaluation-results,2026-09-29,Xiaomi model-card comparison,Same table and named benchmark/version; evaluation harness details are not fully specified.
benchmark,CyberGym,Cybersecurity,reported score points,94.0,95.1,https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL#3-evaluation-results,2026-09-29,Xiaomi model-card comparison,Same table and named benchmark/version; evaluation harness details are not fully specified.
benchmark,ExploitGym,Cybersecurity,reported score points,17.8,6.0,https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL#3-evaluation-results,2026-09-29,Xiaomi model-card comparison,Same table and named benchmark/version; evaluation harness details are not fully specified.
benchmark,ExploitBench,Cybersecurity,reported score points,47.9,25.3,https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL#3-evaluation-results,2026-09-29,Xiaomi model-card comparison,Same table and named benchmark/version; evaluation harness details are not fully specified.
