MiMo-V2.5 mimo-v2.5

xiaomi · mimo

unranked — no quality signals yet

params
310.8B
arch
moe
context
1050k
license
mit open weights
released
Apr 2026
downloads/30d
328.3k

# architecture

attn shared GQA 64q/4kv router top-8 of 256 expert ×256 expert out ×48 layers ctx 1,050,000 310.8B pool
layers
48
d_model
4096
heads
64q / 4kv
head dim
192
experts
top-8 of 256
vocab
152576
family
mimo_v2

signal path of one layer, generated from the registry's structured fields — dimension lines quote the model's real numbers; an MoE trace forks at the router, a dense trace runs straight through. Geometry fields come from the repo's config.json.

# trend

collecting history — trends appear after a few daily runs.

# benchmarks

benchmarkscoresourcedate
Charxiv-r 0.8 / 1 LLM Stats
Claw-eval 0.6 / 1 LLM Stats
LMArena Elo (Code) 1438.0 LMArena Aug 2026
LMArena Elo (Vision) 1238.0 LMArena Aug 2026

# where to run

providerquantctx$/M in$/M out$/M cacheprice srctpsuptime
Featherless via hfrouter
GMICloudfp8 1050k $0.12$0.24$0.00 via openrouter 70.6%
AtlasCloudfp8 1000k $0.14$0.28 atlascloud
Xiaomifp8 1024k $0.14$0.28$0.00 via openrouter 99.1%
Novitafp8 1024k $0.17$0.34 novita 94.9%
Phalafp8 1050k $0.17$0.34 phala
StreamLakeunknown 1000k $0.17$0.34$0.00 via openrouter 96.1%
DeepInfrafp8 256k $0.40$2.00 deepinfra 99.9%
Venicefp8 1000k $0.40$2.00 venice

sorted by blended price ((3·input + output) / 4 per 1M) · ✓ = the provider's own catalog confirms the offer · "via …" prices are what the aggregator routing the offer charges, not the provider's own list price

source aliases
arena
mimo-v2.5
llmstats
mimo-v2.5
openrouter
xiaomi/mimo-v2.5