What is Startlux?
StartLux-Decision is a family of open-weight models that answer typed questions with a calibrated probability for each option. You get instant results in a single forward pass. When the model is unsure, it writes a short thought first, then gives the probabilities. This AutoThink mode keeps answers fast while improving accuracy on hard decisions.
You can choose from eight models, from 0.8B to 27B parameters, in either Instant or AutoThink versions. The models run on standard hardware like one H200, with median request times as low as 52 ms for the 27B Instant model. You can get the weights from GitHub and deploy them yourself, or use the hosted API.
The models lead on the Decision Index 0.3 public benchmark at every size, beating comparable decision models by 7.4 to 16.9 points. The 27B AutoThink model scores 73.37 on the index and 86.9% pair accuracy on DecisionPivot. The website states free unlimited use, so you can test the models without paying.
Startlux features
- Calibrated probabilities: Every option gets a probability that reflects the model's true confidence.
- One-pass inference: Instant models answer in a single forward pass for low latency.
- AutoThink mode: The model writes a short thought only when it is unsure, keeping fast answers for easy cases.
- Open weights: Download all eight models from GitHub and run them on your own infrastructure.
- Multiple sizes: Choose from 0.8B to 27B parameters to balance speed and accuracy.
What you can do with Startlux
- Route customer support tickets to the right team based on predicted intent probabilities.
- Score loan applications by estimating the likelihood of approval or denial.
- Filter content moderation queues by ranking items by probability of violation.
- Automate document triage by classifying each item into categories with confidence scores.
How to get started with Startlux
- Visit the StartLux-Decision website and read the technical report or API reference.
- Download the model weights from GitHub or try the hosted API.
- Deploy the model on your own hardware or use the API to start making decisions.
Tips for better results with Startlux
- Start with the 0.8B Instant model to test your prompts quickly, then scale up to larger sizes only if accuracy on your specific decisions is insufficient.
- Use AutoThink versions of StartLux for hard decisions where you suspect uncertainty, since the model writes a short thought before giving probabilities when unsure.
- Compare probabilities across all options you provide, not just the top one, to catch cases where StartLux is evenly split and needs a human review.
- Download the open weights from GitHub and run StartLux on your own hardware if you need consistent latency or want to avoid any dependence on the hosted API.
- checks
Startlux pricing
Startlux is free to use according to its website.
What to check before you rely on Startlux
- Verify that your decision types match what StartLux was benchmarked on, since the Decision Index covers public subsets and may not reflect your real-world inputs.
- Check the website for details on API rate limits or usage caps, as it states free unlimited use but does not specify any technical constraints.
- Confirm that the output probabilities are calibrated for your domain by testing on a labeled sample, since benchmark scores do not guarantee performance on your data.
Who Startlux is for
Startlux suits developers, data scientists, ML engineers and product teams. If that is not you, the AI research tools below may fit better.
Similar tools compared with Startlux
| Tool | What it is | Pricing |
|---|---|---|
| Startlux | Open decision models with calibrated probabilities | Free |
| OpenSI | AI model comparison and pricing tool | See website |
| ChatPDF | AI chat and summarizer for PDFs | Freemium |
| Consensus | AI search engine for research papers | See website |
| Scite | AI research assistant with smart citations | Free trial |
Startlux FAQ
What is StartLux-Decision?
It is a family of open-weight AI models that answer typed questions by giving a calibrated probability for each option. You can use them for classification, routing, scoring, and other decision tasks.
Is StartLux-Decision free to use?
The website states 'Free unlimited use' for the models. You can get the weights from GitHub and run them yourself, or use the hosted API without paying.
Who is StartLux-Decision for?
It is for developers and data scientists who need fast, accurate decision models. You can deploy it on your own hardware or use the API, and it works for tasks like routing, scoring, and classification.
What is the difference between Instant and AutoThink models?
Instant models answer in one forward pass without any thinking. AutoThink models also answer in one pass when they are sure, but write a short thought first when they are unsure, which can improve accuracy on hard decisions.
Can I run StartLux-Decision on my own hardware or do I need to use the hosted API?
You can do both. StartLux-Decision offers open weights on GitHub, so you can download the models and deploy them on your own infrastructure, such as a single H200. Alternatively, you can use the hosted API for direct access. The website does not specify minimum hardware requirements beyond that.
What does the free plan for StartLux-Decision include?
The StartLux-Decision website states free unlimited use, so you can test the models without paying. This includes access to the hosted API and the open weights. The website does not mention any limitations on request volume or features under the free plan.
Does StartLux-Decision integrate with other tools or platforms?
The StartLux-Decision website does not mention any specific integrations with external tools or platforms. You can use the hosted API or deploy the open weights yourself, which allows you to build custom integrations. For details on integrations, check the API reference or contact support.
