MiniMax-Text-01.
MiniMax Jan 2025 Flagship
Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MMLU | 88.5% | #24 of 127 | Model card | |
| Arena-Hard-Auto | 89.1% | #11 of 39 | Model card | |
| IFEval | 89.1% | #36 of 126 | Model card | |
| DROP | 87.8% | #12 of 40 | Model card | |
| SuperGPQA | 45.1% | #28 of 58 | Official leaderboard | |
| RULER | 91% | #19 of 38 | Model card | |
| SimpleQA | 23.7% | #34 of 59 | Model card | |
| LongBench v2 | 56.5% | #16 of 27 | Model card | |
| MMLU-Pro | 75.7% | #99 of 151 | Model card | |
| MTOB | 51.7 | #2 of 3 | Model card | |
| Chinese SimpleQA | 67.4% | #10 of 13 | Model card | |
| LLM Confabulation Leaderboard | 23.9% | #47 of 58 | Official leaderboard | |
| Creative Writing (MiniMax in-house) | 81.3% | #1 of 1 | Model card | |
| General Assistant (MiniMax in-house) | 73.9% | #1 of 1 | Model card | |
| Hard Capability (MiniMax in-house) | 64.8% | #1 of 1 | Model card | |
| Instruction Following (MiniMax in-house) | 46.3% | #1 of 1 | Model card | |
| Knowledge Q&A (MiniMax in-house) | 78.6% | #1 of 1 | Model card | |
| Long Context (MiniMax in-house) | 93.8% | #1 of 1 | Model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| GSM8K | 94.8% | #15 of 87 | Model card | |
| MATH | 77.4% | #27 of 93 | Model card | |
| Misguided Attention | 36.5% | #28 of 45 | Official leaderboard | |
| GPQA Diamond | 54.4% | #221 of 284 | Model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| HumanEval | 86.9% | #32 of 95 | Model card | |
| MBPP | 71.7% | #39 of 66 | Model card | |
| EvalPlus | 71.7% | #6 of 6 | Model card | |
| Coding (MiniMax in-house) | 90.2% | #1 of 1 | Model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Elimination Game | 3 | #39 of 56 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Safety (MiniMax in-house) | 90.9% | #1 of 1 | Model card |
The benchmarks the lab chose to headline, with the scores it gave.
Every result in the longer documents the lab published with it. 11 of them aren't in the launch post.