XiaomiMiMo/MiMo-V2.6-Pro-RL model card

Fetched 2026-10-18. Weights repo created 2026-09-21 (HF API createdAt), last modified after that, MIT. Card: flagship of the MiMo-V2.6 series, native omnimodal (text, image, video, audio), 1M context. Architecture table: 70 layers (60 SWA / 10 global), hidden 6144, SWA heads 128/8, global heads 128/8, head dims QK/V 192/128, sliding window 128, 384 routed experts with 8 activated, no shared experts, MTP speculative decoder of 5 SWA layers. The card does not state a total or active parameter count; the HF model page reports 1T params. Benchmark table is a vendor claim: Terminal Bench 2.1 89.9, Terminal Bench 4.0 34.9, DeepSWE v1.1 71.9, Toolathlon-Verified 76.9. The HF eval-results widget repeats Terminal-Bench 2.1 89.9 (harborframework/terminal-bench-2.1, verified=false) and DeepSWE 71.9.