Felo API PlatformFelo API Platform
Back to models

Decisions Fast API

Parallel's Decisions Fast model answers named yes/no, multiple-choice, and rubric-scored questions with probabilities, citations, and live web retrieval inside a 32,000-token context.

Parallel: Decisions Fast is a decision model from Parallel Web Systems, released in November 2026, that returns named, typed answers instead of free-form text. A request carries a state, either text or JSON, and up to 100 questions about that same state, with a ceiling of 32,000 tokens per question that includes the shared state. For yes-or-no questions it returns a yes probability; for multiple-choice it returns a choice with per-option probabilities; for scoring questions it returns a level on an ordered rubric. Answers can include per-question refusals, and request-level citations point to the sources used.

The model combines System One classification speed with live web retrieval, which offsets the recency limits of training data. Each answer can draw on the same retrieved evidence across a batch of questions, and typed answers carry citations for verification. It is positioned as a hybrid between a search engine and a decision model, and answers in a few hundred milliseconds for System One classification models. A demonstration request with four checks retrieved 10 sources and returned four probability-bearing answers in 681 milliseconds. It suits fact checks, entity screening, classification tasks, data enrichment, go-to-market workflows, decision routing, and latency-sensitive decisions where current information matters.

Access runs through the OpenRouter Decisions API rather than the OpenAI-compatible chat endpoint, so chat completions SDKs will not work. Per-record latency depends on the concurrency configuration, it is not currently available through EU in-region routing, and the vendor-reported benchmark summary omits datasets, ground-truth process, abstention rate, concurrency settings, and comparison baselines.

Updated:

MODEL ID

decisions-fast
Context
32K
Max output
28.8K
Input
text
Output
decisions

Highlights

  • Typed answers with probabilities and citations
  • Live web retrieval for current information
  • Up to 100 questions per request
  • 32,000-token context
  • Per-question refusal support
  • System One classification speed

Providers

1 providers

Compare API prices across providers. Select a provider to view all charges and conditions.

Provider · Pricing details
Uptime
$0

Benchmarks

  • Know Your Business refresh scenario

    Of 426 public companies, 98 percent of existing flags held up; vendor-reported

    98%
  • Know Your Business refresh scenario

    Seconds; complete batch run elapsed time across 426 public companies; vendor-reported

    <10
  • Lead routing scenario

    Of 500 inbound leads, 97 percent of autonomously routed leads reached the correct segment; vendor-reported

    97%
  • Lead routing scenario

    Seconds; complete batch run elapsed time across 500 inbound leads; vendor-reported

    <9

Scores as published by the model provider.

Decisions Quick Start

Send state and narrow typed questions to /api/v1/decisions and read one answer per question: noul returns a probability from 0 to 1, choice the selected option with its distribution, and score a numeric score for ordered criteria. This endpoint is separate from the Chat Completions, Messages, and Responses surfaces.

curl https://openapi.felo.ai/api/v1/decisions \
  -H "Authorization: Bearer $FELO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "decisions-fast",
    "state": "Help! My payouts have been failing for 3 days.",
    "questions": {
      "is_urgent": {
        "type": "noul",
        "instructions": "Does this message convey urgency?",
        "criteria": {"true": "Time-sensitive", "false": "No urgency"}
      },
      "department": {
        "type": "choice",
        "instructions": "Which team should handle this?",
        "criteria": {
          "billing": "Payments, invoicing, refunds",
          "technical": "Bugs, outages, integrations",
          "sales": "Pricing, upgrades, new accounts"
        }
      },
      "frustration": {
        "type": "score",
        "instructions": "How frustrated is the customer?",
        "criteria": ["Calm", "Frustrated", "Very angry"]
      }
    }
  }'

Frequently asked questions

What is Decisions Fast?

Parallel: Decisions Fast is a decision model from Parallel Web Systems, released in November 2026, that returns named, typed answers instead of free-form text. A request carries a state, either text or JSON, and up to 100 questions about that same state, with a ceiling of 32,000 tokens per question that includes the shared state. For yes-or-no questions it returns a yes probability; for multiple-choice it returns a choice with per-option probabilities; for scoring questions it returns a level on an ordered rubric. Answers can include per-question refusals, and request-level citations point to the sources used. The model combines System One classification speed with live web retrieval, which offsets the recency limits of training data. Each answer can draw on the same retrieved evidence across a batch of questions, and typed answers carry citations for verification. It is positioned as a hybrid between a search engine and a decision model, and answers in a few hundred milliseconds for System One classification models. A demonstration request with four checks retrieved 10 sources and returned four probability-bearing answers in 681 milliseconds. It suits fact checks, entity screening, classification tasks, data enrichment, go-to-market workflows, decision routing, and latency-sensitive decisions where current information matters. Access runs through the OpenRouter Decisions API rather than the OpenAI-compatible chat endpoint, so chat completions SDKs will not work. Per-record latency depends on the concurrency configuration, it is not currently available through EU in-region routing, and the vendor-reported benchmark summary omits datasets, ground-truth process, abstention rate, concurrency settings, and comparison baselines.

How is usage billed?

Your bill depends on the provider you use, its prices and your usage. Longer inputs or other conditions may have different prices.

What kinds of input does it accept?

It reads content passed to it as a state, which can be text or JSON. A single request can carry up to 100 questions about that same state.

What does it return?

It returns typed structured answers: a yes probability for yes-or-no questions, a choice with per-option probabilities for multiple-choice questions, and a level on an ordered rubric for scoring questions. Each answer can include a per-question refusal, and the request returns citations to the sources used.

How much context does it support?

It has a context of 32,000 tokens, and up to 32,000 tokens per question including the shared state.

What work is it designed for?

It suits fact checks, entity screening, classification, data enrichment, go-to-market workflows, decision routing, and other latency-sensitive decisions where current information is critical.

How does it differ from Jev?

In a Jev example, Jev answered no with 0.86 probability while Decisions retrieved current pages and answered yes with probability 1.0, showing the value of live retrieval on questions where the facts have changed.

Can I call it with OpenAI-compatible chat SDKs?

No. It runs on the OpenRouter Decisions API rather than the OpenAI-compatible chat endpoint, so chat completions SDKs will not work with it.

Related models