Model Details
MiMo V2.5 Pro
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 | 78.9% | Benchgen evaluation | 2025-07 |
| GSM8K | 99.6% | Benchgen evaluation | 2025-07 |
| MATH | 86.2% | Benchgen evaluation | 2025-07 |
| MMMU-Pro | 77.9% | Benchgen evaluation | 2025-07 |
| Humanity's Last Exam | 34.0% | Benchgen evaluation | 2025-07 |
| 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 | 80.2% | — | — | Apache 2.0 |
| DeepSeek-R1 0528 | — | 97.3% | 96.2% | MIT |
| 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 · DeepSeek-R1 0528 · Gemini 3 Flash
- Benchmarks: SWE-Bench Verified · MATH · 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.
Benchmark Leaderboards
This model isn’t on any benchmark leaderboard yet.