Model

MiMo-V2.5-Pro

OPEN

Xiaomi · MiMo family · released Apr 22, 2026

Xiaomi's flagship coding/agentic model, 1.02T total / 42B active MoE, 1M context, fully open-sourced under MIT. Optimized for very long agentic trajectories (1,000+ tool calls); unlike sibling MiMo-V2.5, it is text-only (no native image/video/audio encoders).

ReasoningCodingVisionFunction callingTool useAgentic

2199.4

Elo · rank #45

Parameters

1020B

Active params

42B (MoE)

Context

1M tokens

Architecture

Sparse MoE (384 routed experts, 8/token), 70 layers (1 dense + 69 MoE), hybrid Sliding-Window+Global attention 6:1 with learnable attention-sink bias, 3x MTP modules; text-only (no native vision/audio encoders)

License

MIT

Languages

—

API price (in/out)

$0.435 / $0.87

Modalities

text

Benchmark results

Bar shows position within the tracked field; marker = field best

Arena EloHuman preference1466#4

best: Claude Fable 5 · 1505

GDPval-AAAgents1581#10

best: Claude Fable 5 · 1932

GPQA DiamondReasoning86.6%#49

best: GPT-5.6 · 94.6%

Humanity's Last ExamReasoning33.8%#38

best: Claude Sonnet 5 · 57.4%

IFBenchReasoning79.9%#5

best: MiniMax M3 · 83.0%

LiveCodeBenchCoding81.3%#45

best: DeepSeek-V4-Pro (Think Max) · 93.5%

MMLU-ProKnowledge84.6%#31

best: Claude Fable 5 · 91.5%

PinchBenchAgents87.5%#4

best: Trinity-Large-Thinking · 91.9%

SWE-bench ProCoding57.2%#18

best: Claude Fable 5 · 80.0%

SWE-bench VerifiedCoding74.0%#45

best: Claude Fable 5 · 95.0%

τ²-Bench TelecomAgents94.2%#15

best: Claude Opus 4.6 · 99.3%

τ³-BankingAgents8.7%#3

best: Grok 4.5 · 33.0%

Terminal-Bench 2.0Coding57.3%#36

best: GPT-5.6 · 88.8%

Run it locally

VRAM @ Q4

—

VRAM @ FP16

—

Fits on (Q4)

Multi-node cluster required

Throughput data unavailable.

Quantizations

—

MiMo family

Elo progression across releases

MiMo-V2.5Apr 20262077.0 MiMo-V2.5-ProApr 20262199.4

Resources

HFXiaomiMiMo/MiMo-V2.5-Pro DOCShttps://mimo.xiaomi.com/mimo-v2-5-pro/APIAvailable via Xiaomi API

API price $0.435/$0.87 · each benchmark row carries its own source badge (see methodology)