MiMo-V2.5 Pro

Language Xiaomi Verified Retired Shut down Oct 21, 2026 Replaced by MiMo-V2.6 Pro Released Apr 23, 2026

MiMo-V2.5-Pro is Xiaomi’s open-source mixture-of-experts language model intended for demanding agent workflows, complex software engineering, and tasks requiring many steps over an extended sequence. Summary written by AI Index from Xiaomi’s own pages.

  • In Input modalities: Text Out Output modalities: Text
  • Reasoning
  • Reasoning can be off
  • Tool use
  • Structured output
  • Open weights

Prices

2 offerings · prices as listed
Xiaomi MiMo API (opens in a new tab) Official Retired China · CNY mimo-v2.5-pro
Conditions/ 1M tokensInputCache readOutput
Standard no longer listed since Oct 28, 2026 ¥3.00¥0.025¥6.00

Prices from Xiaomi MiMo: pay-as-you-go pricing (opens in a new tab) (checked Oct 19, 2026)

Xiaomi MiMo API (opens in a new tab) Official Retired Global · USD mimo-v2.5-pro
Conditions/ 1M tokensInputCache readOutput
Standard no longer listed since Oct 28, 2026 $0.435$0.0036$0.87

Prices from Xiaomi MiMo: pay-as-you-go pricing (opens in a new tab) (checked Oct 19, 2026)

Standard: the price a request with no conditions pays (the headline price elsewhere on the site).

Details

Knowledge cutoff
Dec 2024
Family
mimo-pro
Model ID

Not stated on Xiaomi’s pages: max input, reasoning efforts, max thinking budget, built-in tools, licence, aliases.

Run locally

Weights: XiaomiMiMo/MiMo-V2.5-Pro
  • Parameters 1.02T
  • Active 42B · 384 experts, 8 per token
  • Native precision BF16
  • Layers 70
  • KV cache per token 420 KB
  • Config positions 1024K (config.json max_position_embeddings)
  • API context 977K (maker’s API)
My hardware GeForce RTX 4090 24 GB · 1008 GB/s · 64 GB DDR5 dual channel · 89.6 GB/s
On your hardware · GeForce RTX 4090
Doesn’t fit IQ4_XS (automatic) 8K context
WeightsKV cacheOverheadUsable
Needs
≈ 534.8 GB (weights 506.2 + KV cache 3.3 + overhead 25.3)
Usable
21.6 GB (90% of its memory)
Try
it’s too large for this hardware; see the smallest that runs it below
PrecisionWeightsNeeds at 8KOn your hardwareLongest context
BF16 (native)maker’s file962.4 GB≈ 1013.8Doesn’t fit—
Q8_01012.5 GB≈ 1066.4Doesn’t fit—
FP8952.9 GB≈ 1003.8Doesn’t fit—
Q6_K782.6 GB≈ 825.0Doesn’t fit—
Q5_K_M677.8 GB≈ 714.9Doesn’t fit—
Q5_0655.1 GB≈ 691.2Doesn’t fit—
Q4_K_M578.9 GB≈ 611.1Doesn’t fit—
Q4_0536.0 GB≈ 566.1Doesn’t fit—
IQ4_XS506.2 GB≈ 534.8Doesn’t fit—
Q3_K_Mquality drops465.7 GB≈ 492.3Doesn’t fit—
IQ3_XXSquality drops364.5 GB≈ 386.0Doesn’t fit—
Q2_Kquality drops357.3 GB≈ 378.5Doesn’t fit—
IQ2_XXSquality drops245.4 GB≈ 260.9Doesn’t fit—

Click a row to choose a precision. “Maker’s file” sizes are the maker’s own files; others are parameters × bits per weight.

Compared with other hardware

Each at the highest precision it runs. Click one for details.

Smallest that runs it (Q4_K_M, 8K): no single listed machine; about 9 × 80 GB GPUs.

Run commands

The maker publishes no files at this precision. Choose its native precision or one of its own builds, or quantise it yourself with llama.cpp.

Memory to fine-tune
Full≈ 15250.3 GBDoesn’t fit your hardware
LoRA≈ 1910.8 GBDoesn’t fit your hardware
QLoRA≈ 541.0 GBDoesn’t fit your hardware

Batch 1, 2,048-token sequences, gradient checkpointing; LoRA and QLoRA train rank-16 adapters on the attention projections.

Estimates; facts from the maker’s Hugging Face repositories, as of 2026-11-21. How it’s estimated

Cited scores

From third-party leaderboards, not measured by us
  • LMArena Text 1,467

    #36 of 163 listed models · CI 1,463–1,470 · 76,826 votes

  • EQ-Bench 4 1,198

    #18 of 31 listed models · CI 1,185–1,213

  • EQ-Bench Creative Writing v3 1,493

    #54 of 95 listed models

  • EQ-Bench Longform Writing 73.7

    #30 of 93 listed models

  • LMArena (Arena), leaderboard-dataset on Hugging Face, licensed CC BY 4.0 · License · as of Nov 8, 2026
  • EQ-Bench by Samuel J. Paech (eqbench.com; EQ-bench/EQ-bench-site on GitHub), MIT License. · License · as of Oct 24 – 28, 2026
  • changes we made
  1. Available → Retired on Xiaomi MiMo API
  2. Available → Retired on Xiaomi MiMo API, China
  3. Deprecated → Retired
  4. New language model

Sources

Xiaomi · China · mimo.xiaomi.com

Snapshots, the same family and other models from this provider