MiniMax-M3 by Nebius Token Factory | AI model information

$10 off the fastest Kimi K3 and GLM 5.3$10 off the fastest Kimi K3 and GLM 5.3

Model details

MiniMax-M3

MiniMax-M3 is a multimodal M-series model from MiniMax, extending the line beyond text-only predecessors with image and video understanding alongside language input. It is documented with a one-million-token that quick-info value window and a mixture-of-experts architecture totaling 427 billion parameters, with about 23 billion activated for a token. This design is intended to support substantial that quick-info value while activating only part of the model’s capacity per step, making it relevant to complex analysis and multimodal work.

In independent testing, MiniMax-M3 scored 55 on the Artificial Analysis Intelligence Index, ahead of the cited open-weight peers, and reached 1,670 on GDPval-AA, matching the cited Claude Sonnet 4.6 result. It also improved substantially over MiniMax-M2.7 on HLE, GPQA Diamond, AA-LCR, IFBench, and CritPt, though SciCode declined slightly. Its practical fit is strongest for broad, real-world tasks spanning many occupations, while available results are less competitive in some coding and agentic evaluations.

Nebius Token FactoryMiniMaxAI/MiniMax-M3minimax

Open provider page Provider docs

Quick Info

Provider: Nebius Token Factory
Model key: MiniMaxAI/MiniMax-M3
Release date: Jun 1, 2026
Last updated: Jun 1, 2026
Input modalities
Output modalities
Capabilities

Cost

Input token cost: $0.30
Output token cost: $1.20

Limits

Output tokens: 1,048,576 tokens
Context window: 1,048,576 tokens

Transparent token rates

Compare MiniMax-M3 pricing

Rates are shown per one million tokens. Combined means one million input plus one million output tokens.

Browse this family

Latest news about MiniMax-M3

GMI Cloud

CoverageAnalysis

Artificial Analysis published a detailed independent evaluation of MiniMax-M3 (article dated June 8, 2026), positioning it as MiniMax's first multimodal M-series model with a 1M-token context window and native vision input (image and video), succeeding the text-only MiniMax-M2.7. The model scores 55 on the Artificial A The article documents specific capability gains over MiniMax-M2.7: HLE +9 points (28% to 37%), GPQA Diamond +6 (87% to 93%), AA-LCR +5 (69% to 74%), IFBench +7 (76% to 83%), and CritPt +3 (1% to 4%), with a small SciCode regression (47% to 45%). It scores 1670 on GDPval-AA (level with Claude Sonnet 4.6 max), 80% on MMM