MiMo-V2.6 Technical ReportProvider report
2026-09-22Complete Table 3 on page 26 of the MiMo-V2.6 technical report: 17 metrics and 90 scores. Missing cells do not create scores. Final evaluations are kept separate from Live RL training curves. Configurable baselines use maximum reasoning effort. MiMo Code Bench, MiMo Cyber Bench and MiMo Visual Coding are internal benchmarks; CyberGym uses corrected evaluation environments. MiMo-V2.5-Pro has no OSWorld result in the PDF or model card; the release image attributes 61.5 to MiMo-V2.5, not Pro.