Model

MiniMax-VL-01

OPEN

MiniMax · MiniMax-01 family · released Jan 14, 2025

Vision-language sibling of MiniMax-Text-01, adding a dynamic-resolution (336px-2016px) ViT+MLP encoder atop the same 456B hybrid-attention MoE backbone.

ReasoningCodingVisionFunction callingTool useAgentic

1276.1

Elo · rank #162

Parameters

456B

Active params

45.9B (MoE)

Context

1M tokens

Architecture

Hybrid Lightning+Softmax Attention MoE (80 layers, 32 experts top-2) + 303M-param ViT/MLP vision encoder

License

Custom MiniMax Model License (free for most use; platforms with >100M MAU must request a separate license; code under MIT)

Languages

—

API price (in/out)

$0.2 / $1.1

Modalities

text · vision

Benchmark results

Bar shows position within the tracked field; marker = field best

AI2DVision83.3%#30

best: Molmo 72B · 96.3%

ChartQAVision91.7%#1

best: this model · 91.7%

DocVQAVision96.4%#2

best: Qwen2-VL-72B · 96.5%

MathVistaVision68.6%#31

best: Seed 2.1 Pro · 90.7%

MMMU-ProVision52.7%#42

best: Claude Opus 4.7 · 85.5%

MMMUVision68.5%#60

best: Claude Fable 5 · 89.3%

OCRBenchVision865#7

best: InternVL3-78B · 906

OlympiadBenchMath24.2%#4

best: Doubao-1.5-Pro · 59.8%

Run it locally

VRAM @ Q4

—

VRAM @ FP16

912 GB

Fits on (Q4)

Multi-node cluster required

Throughput data unavailable.

Quantizations

—

MiniMax-01 family

Elo progression across releases

MiniMax-VL-01Jan 20251276.1 MiniMax-Text-01Jan 20251288.1

Resources

HFMiniMaxAI/MiniMax-VL-01 GHMiniMax-AI/MiniMax-01 DOCSplatform.minimax.io APIAvailable via MiniMax API

API price $0.2/$1.1 · each benchmark row carries its own source badge (see methodology)