Perplexity's open-weight decision model that picks among given options for routing and checks in AI systems
LimiX-2
A 400M-parameter model for tables that does classification, regression and missing-value filling, under a non-commercial licence
About LimiX-2
LimiX-2 works on spreadsheets and database tables rather than on prose, which makes it unusual among the models in the AI Model category. It came out on 16 September 2026 from Stable AI, working with a research group at Tsinghua University led by Professor Peng Cui.
The licence comes first
The weights can be downloaded, but they carry the licence named stableai-limix-non-commercial-license-v1.0, which allows non-commercial use only. It is therefore not open source, and a company cannot freely build a commercial product on it. A business should settle that before looking at anything else about it.
One checkpoint for three table tasks
A single pretrained checkpoint of 400 million parameters handles classification, regression and the filling in of missing values across tables it has never seen, without task-specific updates to its parameters. Those jobs are normally given to a gradient-boosting library trained afresh for each dataset. The authors call the architecture a Contextual Mechanism Network, and its pretraining used Context-Conditional Masked Modeling, applied to synthetic datasets that were produced from structural causal models. The authors add that the attention the model pays to features captures direct causal links closely enough to reconstruct a causal skeleton. That is a claim about explanation, not accuracy.
Leaderboard results as the vendor reports them
Stable AI says it ranked first on all three tabular leaderboards it was evaluated on: an Elo of 1935 on TabArena, 1506 on TALENT and 1432 on BCCO, each ahead of the runner-up. Those figures come from its own model card and paper. They are Elo ratings on leaderboards that no other model in the directory has been measured against, so they are not stored as comparable scores.
Availability
Code for running the model, along with the checkpoint, is hosted on GitHub and Hugging Face.
TrustList summary
LimiX-2 is a tabular model built for structured data rather than text. Stable AI and Professor Peng Cui's group at Tsinghua University put it out on 16 September 2026. Its licence is the main thing a business needs to check, since it allows non-commercial use only and is not open source.
What LimiX-2 does
- Classification: sorts table rows into classes across tables the model has never seen, without task-specific training.
- Regression: predicts numeric values from table columns.
- Missing-value imputation: fills gaps in tables from one pretrained checkpoint.
- Causal attention: the authors say its feature attention can recover a direct causal skeleton, a claim about explanation rather than accuracy.
Clients
Data teams that work with tables and can use the model under a non-commercial licence.
Key facts
- Size: a 400M-parameter checkpoint.
- Licence: a non-commercial licence, stableai-limix-non-commercial-license-v1.0.
- Code: inference code and the model checkpoint are available on GitHub and Hugging Face.
Summary written by TrustList from the business's own published information, Nov 2026.
Benchmarks & AI stats
| Benchmark | Official | Community avg |
|---|---|---|
| No benchmark scores yet. Be the first to add one. | ||
“Official” values are editor-approved and feed the ranking. “Community avg” is the mean of member submissions (shown for transparency; it never affects the ranking until an editor approves a value).
Sign in to add a benchmark score for this model.
LimiX-2 launches
Request a demo or quote from LimiX-2
Reviews
Questions & answers
No questions yet. Be the first to ask about LimiX-2.
LimiX-2 alternatives
Similar listings buyers compare against this one.
Open-source 2B decision model from AWS's Strands Agents project that picks options and returns scores with a reliability figure
Liquid AI's open-weight decision model, giving yes or no, choice or score answers from text and images
Black Forest Labs' image generation and editing model with bounding-box layouts, multiple reference images and 4K output
Aleph Alpha's open-weight English-German mixture-of-experts model for self-hosting, released under Apache 2.0 in October 2026
JetBrains' open 12B mixture-of-experts coding model with 2.5B parameters active per token under Apache 2.0
Google's open multimodal embedding model that maps text, images, audio and video into one vector space, under Apache 2.0
Cantina Security's first open-weight security model, a post-trained GLM-5.3-Flash for vulnerability research and exploit checks
More on TrustList
Everything here links back to the same verified catalogue. Pick your next stop.
- Write a reviewHelp other buyers decide
- CompaniesAgencies, consultancies and IT service providers, ranked by verified reviews.
- ProductsSoftware and SaaS with pricing, features, integrations and alternatives.
- ArticlesGuides, comparisons and research from the editorial desk and the community.
- AwardsAnnual recognition decided by verified reviews and an independent jury.
- LaunchesNew products and releases, voted up by the community every day.
- RequestsBuyers describe what they need; vendors respond directly.
- PeopleReviewers, authors and makers with public profiles.
- ComparePut up to four listings side by side before you shortlist.