structured decision
AI Decision Models 2027
Every structured decision model Felo carries for 2027 in one place: typed answers, live token rates and the same key as chat models.
- 2027 lineup
- All providers in one key
- Typed answers, not text
7 models · 7 Providers
The 2027 lineup
7 models
2027 picks
Morph: ReflexNewMorph: Reflex is a specialist model that classifies every turn in an agent loop, returning one deterministic answer from predefined options without generating text.$0.042/ 1M tokens$0.01/ 1M tokensPricingPOST /api/v1/decisionsModalities: Text · Image → decisions02
Cloudflare: ClairicNewClairic is a Cloudflare extraction model that pulls strongly typed values from text and images in one pass, reporting confidence per field.$0.1/ 1M tokensPricingPOST /api/v1/decisionsModalities: Text · Image → decisions03
Baseten: Joey 1NewBaseten's Joey 1 is a 2B-parameter open-weight decision model that returns calibrated probabilities for predefined choices instead of generating text, built for fast, high-volume classification and routing.$0.01/ 1M tokensPricingPOST /api/v1/decisionsModalities: Text · Image → decisions04
LiquidAI: d1.1NewLiquidAI's d1.1 is a decision model that returns probabilities for yes/no, choice, and score questions instead of generating text.$0.02/ 1M tokens$0.02/ 1M tokensPricingPOST /api/v1/decisionsModalities: Text · Image → decisions05
Nace.AI: Drex MaxNewNace.AI: Drex Max is a multimodal decision model that returns calibrated probabilities for typed questions over text, JSON and images, with a 262,144-token context and 183-millisecond latency.$0.03/ 1M tokens$0.03/ 1M tokensPricingPOST /api/v1/decisionsModalities: Text · Image → decisions06
Perplexity: Decider Flash V1.2NewPerplexity's Decider Flash V1.2 is a 4B-parameter, open-weight decision model built on Qwen3.5-4B that returns probabilities instead of prose for classification and scoring tasks.$0.01/ 1M tokensPricingPOST /api/v1/decisionsModalities: Text · Image → decisions07
TypeSafe: Jev 1.14NewTypeSafe: Jev 1.14 returns typed decision outputs for model routing and classification, with experimental image and long-context support.$0.042/ 1M tokensPricingPOST /api/v1/decisionsModalities: Text · Image → decisionsWhat you can build
Request routing
Send each request to the right model, queue or team by letting a classifier choose between options.
Classification and moderation
Label content into a fixed set of categories and get the confidence with the label.
Ranking and prioritisation
Order tickets, leads or candidates by an explicit score instead of a paragraph of reasoning.
Approval gates
Decide whether an action may proceed, and keep the probability for audits.
How it works
- 1
Define the decision
Write the question, the options or the scoring scale, and the context the model may use.
- 2
Send it to decisions
POST to /api/v1/decisions with your Felo key and a decision model ID.
- 3
Read the typed answer
The response carries the chosen option, probability or score; your code branches on that value.
curl https://openapi.felo.ai/api/v1/decisions \
-H "Authorization: Bearer $FELO_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"reflex","input":"Route this request to one of the options.","options":["support","sales","billing"]}'
How billing works
Decision routes are billed per token. Some models only bill input tokens because the answer is a label or a probability; the table shows the rate and the unit for each model.
- Rates come from the default channel and are deducted from prepaid USD credits.
- Top up from $1; there is no subscription and no monthly minimum.
- Models that bill input only say so on their model page; check it before sizing volume.
Try it online, no code
Test any model in the Playground
Pick a model, send a request and read the response in the browser. No local setup, no install.
Open PlaygroundModel cases
Browse all model casesReal outputs and prompts from models in this lineup.
Why teams use Felo
Felo carries the structured decision models available in 2027 — TypeSafe, Baseten, Cloudflare, Liquid AI, Nace.AI and Perplexity — under one key and one bill, on the same endpoint as each other.
- Every model in the 2027 lineup is callable with the same key and the same endpoint.
- Compare decision models side by side instead of testing them one vendor at a time.
- The same key and balance cover chat, image and video models too.
Frequently asked questions
What is a decision model API?
A decision model API answers a constrained question with a typed result: a probability, a chosen option with its distribution, or an ordered score. It is built for routing, classification and approval steps where parsing prose would be fragile.
How is a decision model different from a chat model?
A chat model generates text. A decision model returns a machine-readable answer to a defined question, so your code can branch on it without prompt engineering the output shape or parsing free text.
Which decision models are available on Felo?
The catalog currently lists structured decision models from TypeSafe, Baseten, Cloudflare, Liquid AI, Nace.AI and Perplexity. The table shows each model's provider, rate and response shape.
How are decision models billed?
Per token. Some models bill input tokens only, because the output is a label or a probability; the model page lists the exact rate and unit, and calls are deducted from prepaid USD credits.
Can I try a decision model without writing code?
Yes. Open the Playground to select a decision model, send a request and read the typed response. The same models are documented in the Data API reference.
Can I use decision models for routing inside an agent?
Yes. They are designed for that: give the model a fixed set of options and use the returned choice or score to decide the next tool, model or queue. The endpoint is POST /api/v1/decisions with the same key as the other models.
What are the best decision models in 2027?
Felo lists 7 structured decision models from 7 providers, including TypeSafe, Baseten, Cloudflare, Liquid AI, Nace.AI and Perplexity; the best pick depends on the answer shape you need.
Call your first decision model
Create a key, choose a decision model ID and send one request to POST /api/v1/decisions. The Playground is there if you would rather click than code.