MiMo V2.5

Released 2026-04 · multimodal · 1.0M tokens · 2 benchmarks · Open weight

Editorial notes

Lanzado 22 abril 2026. Omnimodal (text/image/video/audio). 311B total / 15B activos MoE. Hybrid attention 5:1, KV-cache reducido ~6x. 100-150 tok/s. Backbone V2-Flash + encoders dedicados de vision y audio. Open-source. Pricing $0.40/$2.00 per MTok. Scores corregidos contra HF card (catalogo previo tenia SWE-Pro/TB2 swapped: 76.0→56.1, 56.1→65.8).

Spec sheet

Company: Xiaomi
Country: CN
Type: multimodal
Release: 2026-04
Context: 1.0M tokens
Params total: 310B
Params active (MoE): 15B
License: mit
Quants: FP8, BF16, Q8_0, Q4_K_M
Pricing (openrouter): $0.4/$2/M
Slug: mimo-v2-5

Quick install

1 tools

docs.vllm.ai

vllm serve XiaomiMiMo/MiMo-V2.5

Benchmarks (2)

Coding 2

Cite this model

BibTeX · APA

BibTeX

@misc{frontier-mimo-v2-5,
  title  = {MiMo V2.5},
  author = {{Xiaomi}},
  year   = {2026},
  note   = {Frontier Benchmarks AI atlas. Accessed 2026-06-17},
  url    = {https://frontierbenchmarks.com/models/mimo-v2-5}
}

APA

Xiaomi (2026). MiMo V2.5 [Large language model]. Frontier Benchmarks AI. Retrieved 2026-06-17, from https://frontierbenchmarks.com/models/mimo-v2-5

Citation reflects the atlas page, not the original model paper. For the paper, see the "Resources" section above.

Battle vs another model ← All models More from Xiaomi Methodology