# MiMo V2.5 Pro ![Organization](https://img.shields.io/badge/Org-Xiaomi-ff6900) ![License](https://img.shields.io/badge/License-Apache%202.0-22c55e) ![Released](https://img.shields.io/badge/Released-Jul%202025-555) ![Reasoning](https://img.shields.io/badge/Type-Reasoning-8250df) > **Quick answer:** MiMo V2.5 Pro is Xiaomi's July 2025 reasoning model scoring 78.9% SWE-Bench Verified, 99.6% GSM8K, 86.2% MATH, 77.9% MMMU-Pro, and 34.0% HLE. Apache 2.0 — Xiaomi's open-source reasoning model with near-perfect math. ## At a Glance **Where MiMo V2.5 Pro leads** - 78.9% SWE-Bench Verified — competitive; near Gemini 3 Flash (78%) and above GPT-5 Codex (74.5%) - 99.6% GSM8K — near-perfect grade-school math - 86.2% MATH — strong competition math - 77.9% MMMU-Pro — strong multimodal understanding benchmark - Apache 2.0 — fully open **Where it lags** - 34.0% HLE — limited frontier reasoning - 68.5% MMLU-Pro — below competing models at this tier - Xiaomi brand: smaller ecosystem than Meta/Alibaba/Google **Best for:** Open-source math and reasoning pipelines; SWE-Bench class coding; Apache 2.0 deployments needing balanced math+coding. ## What MiMo V2.5 Pro Is MiMo V2.5 Pro is Xiaomi's July 2025 reasoning model from the MiMo (Mini Model) series — Xiaomi's AI research team's open-source model line. The V2.5 Pro is designed with a strong math-reasoning focus while also achieving competitive SWE-Bench scores. The 99.6% GSM8K is near-perfect and among the highest reported on this benchmark. The 86.2% MATH is also excellent. The model appears to use chain-of-thought reasoning training similar to DeepSeek-R1, focusing on mathematical and code reasoning. ## Specifications | Field | Value | |---|---| | Organization | Xiaomi | | License | Apache 2.0 | | Release date | July 2025 | | Modality | Text only | ## Pricing Open weights under Apache 2.0 — self-host at no cost. ## Public Benchmark Scores | Benchmark | Score | Source | Date | |---|---|---|---| | [SWE-Bench Verified](/benchmarks/swe-bench-verified) | 78.9% | Benchgen evaluation | 2025-07 | | [GSM8K](/benchmarks/gsm8k) | 99.6% | Benchgen evaluation | 2025-07 | | [MATH](/benchmarks/math) | 86.2% | Benchgen evaluation | 2025-07 | | [MMMU-Pro](/benchmarks/mmmu-pro) | 77.9% | Benchgen evaluation | 2025-07 | | [Humanity's Last Exam](/benchmarks/humanitys-last-exam) | 34.0% | Benchgen evaluation | 2025-07 | | [MMLU-Pro](/benchmarks/mmlu-pro) | 68.5% | Benchgen evaluation | 2025-07 | ## MiMo V2.5 Pro vs Alternatives | Model | SWE-Bench | MATH | GSM8K | License | |---|---|---|---|---| | **MiMo V2.5 Pro** | 78.9% | 86.2% | 99.6% | Apache 2.0 | | [Kimi K2.6](/models/moonshot/kimi-k2-6) | 80.2% | — | — | Apache 2.0 | | [DeepSeek-R1 0528](/models/deepseek/deepseek-r1-0528) | — | 97.3% | 96.2% | MIT | | [Gemini 3 Flash](/models/google/gemini-3-flash) | 78% | 97.5% | 96.8% | Proprietary | MiMo V2.5 Pro leads on GSM8K (99.6%) among Apache 2.0 models. SWE-Bench is competitive (78.9% — near Kimi K2.6's 80.2%). For open-source math-heavy pipelines with SWE-Bench performance: MiMo V2.5 Pro is strong. ## Related Models & Benchmarks - Compare: [Kimi K2.6](/models/moonshot/kimi-k2-6) · [DeepSeek-R1 0528](/models/deepseek/deepseek-r1-0528) · [Gemini 3 Flash](/models/google/gemini-3-flash) - Benchmarks: [SWE-Bench Verified](/benchmarks/swe-bench-verified) · [MATH](/benchmarks/math) · [GSM8K](/benchmarks/gsm8k) ## Frequently Asked Questions

What is MiMo V2.5 Pro?

Xiaomi's July 2025 reasoning model scoring 78.9% SWE-Bench Verified, 99.6% GSM8K, 86.2% MATH, 77.9% MMMU-Pro. Apache 2.0.

Who makes MiMo?

MiMo is developed by Xiaomi's AI research team. The MiMo series (Mini Model) represents Xiaomi's open-source LLM research, focused on reasoning and mathematical tasks.

--- *Specs from Xiaomi's MiMo V2.5 Pro release (July 2025) and Benchgen evaluations. Last updated 2026-07-24.*

MiMo-V2.5-Pro — Evaluations

Evaluations