Quick answer: MiniMax M2 is MiniMax's August 2025 model scoring 83.0% LiveCodeBench, 82% MMLU-Pro, 44.0% BrowseComp, and 36.0% SciCode. MiniMax Commercial License — notably high LiveCodeBench at 83.0%.
Where MiniMax M2 leads
Where it lags
Best for: LiveCodeBench-class coding deployments; MiniMax ecosystem integrations; balanced coding + knowledge tasks.
MiniMax M2 is the August 2025 model from MiniMax — a Chinese AI company known for multimodal models and the MiniMax commercial API. The M2 represents their coding and knowledge-focused general model.
The 83.0% LiveCodeBench is the standout metric — one of the highest scores publicly reported on this benchmark, competitive with NVIDIA Nemotron 3 Super 120B (81.2%) and DeepSeek-V3.2 Exp (74.1%). This makes MiniMax M2 a compelling option for coding-agent pipelines.
| Field | Value |
|---|---|
| Organization | MiniMax |
| License | MiniMax Commercial License |
| Release date | August 2025 |
| Modality | Text only |
Available via MiniMax API. Refer to MiniMax pricing for current M2 rates.
| Benchmark | Score | Source | Date |
|---|---|---|---|
| LiveCodeBench | 83.0% | Benchgen evaluation | 2025-08 |
| MMLU-Pro | 82% | Benchgen evaluation | 2025-08 |
| BrowseComp | 44.0% | Benchgen evaluation | 2025-08 |
| SciCode | 36.0% | Benchgen evaluation | 2025-08 |
| Model | LiveCodeBench | MMLU-Pro | License |
|---|---|---|---|
| MiniMax M2 | 83.0% | 82% | MiniMax Commercial |
| Nemotron 3 Super 120B | 81.2% | 83.7% | NVIDIA Open |
| DeepSeek-V3.2 Exp | 74.1% | — | MIT |
| Qwen3 235B A22B | 65.9% | 68.2% | Apache 2.0 |
MiniMax M2 leads on LiveCodeBench (83.0% vs 81.2% Nemotron 3 Super). For open-source alternatives with high LiveCodeBench: Nemotron 3 Super 120B (NVIDIA Open) is the closest.
Specs from MiniMax's M2 release (August 2025) and Benchgen evaluations. Last updated 2026-07-24.
This model isn’t on any benchmark leaderboard yet.