BWhat changed
Scaled a hybrid linear/softmax attention expert-model architecture with optimized parallel execution and communication overlap, releasing text and vision-language models for long-context tasks.
Scientific scope
Scaled a hybrid linear/softmax attention expert-model architecture with optimized parallel execution and communication overlap, releasing text and vision-language models for long-context tasks. Integrated system builds on prior Lightning Attention and MoE methods. Synthetic retrieval or context extrapolation does not imply reliable reasoning over all four-million-token inputs. Text/VL and different context settings form one contribution.
Evidence
The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Integrated system builds on prior Lightning Attention and MoE methods. Synthetic retrieval or context extrapolation does not imply reliable reasoning over all four-million-token inputs. Text/VL and different context settings form one contribution.
Contributors
- Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.
- MiniMax-01: Scaling Foundation Models with Lightning Attention
- MiniMax: Lightning-Attention long-context expert system — official company record
Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.
Full scientific record · Browse contribution grades · Request a correction