Decision models classify, label, rerank and approve in milliseconds, for a fraction of a cent. The open question is whether you can trust their calls.
See how they score on public benchmarks, with speed and cost side by side, in the Decision Leaderboard. Then make the calls yourself in the Decision Arena and find the model that decides like you.
//Use Cases
Reach for a decision model when the same question comes up again and again, for every ticket, document, transaction, or search result, and the answer is a pick from a set, a score, or a yes/no. These four jobs cover most of it.
Sort inputs by intent, category, or meaning, without relying on rigid rules.
Use it when each input belongs in exactly one bucket, like routing a ticket to the right team.
Apply consistent labels to text and other supported content, at the volume your data arrives.
Use it when an input can carry several tags at once, like tagging content for search or analytics.
Judge inputs against your criteria and policies, and get back a verdict you can act on automatically.
Use it when you need a pass/fail or a score against your own rules, like flagging transactions for review.
Order candidates by relevance to a query, so the best results reach your users and your agents first.
Use it when you already have candidates and need the best ones first, like search results or documents for an agent.
//Decision Models
Models that read a state and return a choice, a score, or a yes/no, instead of text. Call any of them through the same API.
25 models14 providers
Baseten
Cloudflare
Inception
Jared Palmer
Liquid AI
Microsoft
Nace.AI
OpenAI
Perplexity
Respan
ScaleDown
Together AI
TypeSafe
Upstage
//Customization
Start with pretrained models and natural-language instructions.
When your requirements become more specific, adapt supported models with domain examples, labeled decisions, and feedback.
Post-training makes it possible to improve task-specific performance without building a foundation model from scratch.
Recurrence post-trains models for specific verticals and decision types, so they start closer to how your industry already decides.
Post-trained for your vertical
01.
Select an available model and describe the task.
02.
Test the model against examples relevant to your domain.
03.
Use supported post-training workflows to improve model behavior.
04.
Run your specialized model on Recurrence or in your own environment.
//Deployment
Every model on Recurrence, including the ones you post-train, runs two ways: hosted by us, or inside your own infrastructure.
Same models, same API. You choose where they run.
Managed
Send requests to Recurrence and get decisions back.
Self-hosted
Deploy the same models in your own cloud account or data center.
Same API
One request format on Recurrence or in your environment. Moving is a config change.
Your models travel
Models you post-train can run either way, not just the stock ones.
Mix per workload
Keep sensitive workloads in-house and send the rest to Recurrence.
Access leading decision models, customize them for your domain, and run them wherever your data lives.
