Multilingual Retrieval Breakthrough: 307M Parameter Model Pushes Pareto Frontier
IgorCarron · x · 2026-07-31
Following the release of their SOTA English models DenseOn and LateOn, the team has extended their capabilities to the multilingual domain.
Experiments reveal that while mDenseOn is a strong model, the performance gap between late interaction (LI) and dense models remains massive on long contexts (like MLDR) and languages outside the training setup. This highlights the strong generalization capabilities of LI models. Furthermore, the new models push the Pareto frontier at 307M parameters across benchmarks including BEIR, MIRACL, MLDR, and MTEB Code.
More from Models
- Google DeepMind Launches Gemini Robotics 2 for Whole-Body Control and Multi-Robot Collaboration — CyberRobooo · 2026-07-31
- DeepMind's Gemini Robotics 2 Enables Multi-Robot Collaboration — CyberRobooo · 2026-07-31
- Claude 4.6 Sonnet Remains the Best Model for Writing Despite Coding Focus — vista8 · 2026-07-31
- LG Releases K-EXAONE 2.0: A 750B Open-Source Model Outperforming GLM-5.1 in Long Context — AlphaLemonMint · 2026-07-31
- AI Search Transparency Critiqued: Should Underlying Indexes Be Public? — iamtrask · 2026-07-31
- Affine Shifts Strategy: Enhancing Software Engineering Capabilities on Top of Frontier Models — markjeffrey · 2026-07-31