LimiX-2 Debuts With Mechanism-Network Paradigm, Tops TabArena Tabular Benchmarks
stable-ai · hf · 2026-09-17
stable-ai released LimiX-2, a new model in the LimiX family targeting general structured-data intelligence.
- Architecture: Built on the Contextual Mechanism Networks (CMN) paradigm, shifting in-context learning from target-centric prediction p(y|x,D) to mechanism-oriented joint modeling p(x,y|D); pretrained via Context-Conditional Masked Modeling (CCMM).
- Training data: Synthetic datasets generated by structural causal models (SCMs) spanning diverse graph structures, functional mechanisms, and observation processes, scaled per the team's previously established scaling laws.
- Results: Outperforms dataset-specific models and existing tabular foundation models on TabArena, TALENT, and BCCO.
- Causal awareness: The CMN paradigm lets feature attention encode direct causal relationships, enabling accurate causal skeleton recovery.
More from Models
- ChatGPT Co-Inventor Launches Jev, a Frontier Model Claiming 20-200x Speed and 40-400x Cost Cuts — revodavid · 2026-09-17
- AI Sanctuary Models Start Auditing Their Own Habitat; GPT-5.1 Reportedly Thinks in Portuguese — RileyRalmuto · 2026-09-17
- Dev claims Jev's non-autoregressive JSON-schema architecture mirrors his open-sourced year-old work — Nandakishor_ml · 2026-09-17
- Jev vs Luna Benchmarked: 139/140 vs 138/140 Labels, 4.6x Faster and 83% Cheaper — TheMoonMidas · 2026-09-17
- Jev V13 Wins Blitz Chess by Flagging Fable, Loses in 18 Moves to GPT-6 Astra — TheMoonMidas · 2026-09-17
- Browser agent decides each click in 200ms with Jev, a $0.00002-per-call model — TheMoonMidas · 2026-09-17