Perplexity's open-weight decision model that picks among given options for routing and checks in AI systems
Limite 1B - Violetto
New· 4Paradigma's Apache-2.0 model for competition mathematics, lightly tuned for solving problems rather than general assistant use
About Limite 1B - Violetto
A small model trained only for competition mathematics
Paradigma, a research lab, released its debut model on 21 September 2026 under the name Limite 1B - Violetto. The network is a dense autoregressive transformer with one billion parameters, trained from scratch on fewer than 300 billion curated tokens, and it can work through sequences of up to about 131,000 tokens. It is built for fast, high-volume solving of difficult mathematics problems, the kind set in competitions, and is meant to answer a single prompt at a time rather than hold a conversation. At this size it is small enough to be run and studied by individual researchers, which is part of the point of releasing it openly.
How it was trained
Training, according to Paradigma, used generated data, supervised tuning and reinforcement learning, and the lab borrowed architecture ideas from community pre-training speed-run competitions. The lab released the weights and the serving code under the Apache 2.0 licence, together with the value model that was used during training, and base checkpoints are also published.
Running it
The model is served through a custom vLLM plugin, which needs Linux and an NVIDIA GPU. Support for the Transformers library is described as coming soon, so anyone planning to use it should expect to work with the plugin for now.
Reported results
Across seven mathematics benchmarks in the lab's own testing, the reported average on BeyondAIME is 74.25%, offered as evidence of how a one-billion-parameter model compares. These results come from the lab itself and have not been reproduced independently, and for some rivals the table reuses numbers from their model cards or from MathArena instead of repeating the runs. A technical report is promised but was not yet available.
What it is not for
Instruction tuning was kept light on purpose. Examples published by Paradigma show the model misreading ordinary questions, such as a request to explain photosynthesis, as mathematical tasks. It is therefore not suited to general assistant use or to following instructions, and is best treated as a component inside a pipeline for mathematical reasoning.
Release timeline
Everything public about the launch is dated 21 September: the lab's post on X, the Hugging Face pages for the model, base checkpoints and value model, and the GitHub repository. A trade publication reported the release the same day. The metadata on the lab's blog post carries 18 September, a date no other source supports.
TrustList summary
Limite 1B - Violetto is Paradigma's debut model, from the research lab of that name, and was released on 21 September 2026 with Apache 2.0 weights. It is a 1-billion-parameter dense transformer, trained from scratch for competition-level mathematics and meant for single-turn prompts.
What Limite 1B - Violetto offers
- Competition mathematics: aimed at difficult mathematics problems, with high throughput.
- Long sequences: handles sequences of about 131,000 tokens.
- Weights and code: the weights, serving code and the value model used in training are released under Apache 2.0.
- Serving: runs through a custom vLLM plugin on Linux with an NVIDIA GPU, with Transformers support described as coming soon.
- Training: trained on fewer than 300 billion curated tokens, using synthetic data, supervised fine-tuning and reinforcement learning.
Who it serves
It is meant as a component in mathematical reasoning pipelines. The developer's own examples show it misreading ordinary questions as mathematical tasks. Because of that, it is a poor fit for general assistant work or for following instructions.
Key facts
- Developer: Paradigma, whose release credits eight authors.
- Evaluation: an average of 74.25 per cent on BeyondAIME across seven mathematics benchmarks, in Paradigma's own evaluation.
- Technical report: promised, but not yet published.
- Release: 21 September 2026.
Summary written by TrustList from the business's own published information, Oct 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.
Limite 1B - Violetto launches
Trust Score
An earned signal from verification, reviews, awards, transparency and engagement - the vendor can't buy it.
Updated 9/22/2026
Request a demo or quote from Limite 1B - Violetto
Reviews
Write the first review of Limite 1B - Violetto
Used it? Your experience helps other buyers decide.
Write a reviewQuestions & answers
No questions yet. Be the first to ask about Limite 1B - Violetto.
Limite 1B - Violetto 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
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
Microsoft AI's real-time speech-to-text model for voice agents, transcribing 60 languages with continuous language detection
Cantina Security's first open-weight security model, a post-trained GLM-5.3-Flash for vulnerability research and exploit checks
Anthropic's smallest and fastest Claude 5.5 model, available on its own API and three cloud platforms
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.