MiMo-V2.5-Pro

Training compute

6.8×10²⁴ FLOP

Parameters

1T

Published

Apr 23, 2026

MiMo-V2.5-Pro is an AI model developed by Xiaomi Corp (China), first published in April 2026. It works in the language domain, on tasks such as language modeling/generation and question answering.

Training it took an estimated 6.8×10²⁴ FLOP of compute. The model has 1,020,000,000,000 parameters. It was trained on roughly 27T datapoints.

Access: Open weights (unrestricted). Its weights are openly available. Epoch AI rates the confidence of this record as likely.

Full record

Organization: Xiaomi Corp
Country of organization: China
Domain: Language
Task: Language modeling/generation, Question answering
Training compute: 6.8×10²⁴ FLOP
Parameters: 1,020,000,000,000
Dataset size: 27T
Numerical format: FP8
Model accessibility: Open weights (unrestricted)
Open weights: Yes
Epoch confidence: Likely

Benchmark results

01	WebDev Arena	Arena Score	1471.3
02	ALE Bench	Performance	899.8
03	SciCode	Score	0.5
04	Proofbench	Accuracy	0.24
05	Critpt	Accuracy	0.04