What is Jev? The $0.042/M "System One" Model & The Jevons Paradox in AI Costs

1. What Is Jev? The System One Shift

In cognitive psychology, System 1 thinking is fast, instinctive, and automatic, while System 2 is slow, deliberative, and analytical.

For the past four years, the AI ecosystem has been dominated almost exclusively by System 2 models: general-purpose LLMs (such as GPT-4o, Claude 3.7 Sonnet, and DeepSeek V4) that process tokens autoregressively to generate long sentences and complex prose.

However, software applications frequently do not need prose. When an AI agent decides which tool to call, whether an email is urgent, or which route an API request should follow, generating JSON text through a 200-billion parameter model is wasteful.

Jev by TypeSafe AI is designed specifically as a System One Decision Model. Given an application state and a question, Jev skips text generation entirely and outputs calibrated numerical choices, probabilities, and classifications directly with sub-100ms latency.

2. The Pricing Breakthrough: $0.042 Input and Free Output

TypeSafe AI's pricing structure has made headlines because it introduces the first commercial foundation model with 100% free output tokens:

Why can TypeSafe AI afford free outputs? In traditional generative AI, output tokens are 3× to 5× more expensive to serve than inputs because generation requires sequential autoregressive decoding. Jev emits only fixed logits and calibrated probabilities. The computational cost of the output is mathematically negligible — literally "too cheap to meter."

3. The "Jevons Paradox": Why Cheaper Tokens Make AI Bills Skyrocket

The model is named directly after William Stanley Jevons, the British economist who in 1865 observed that as James Watt's steam engines made coal consumption vastly more efficient, the total consumption of coal skyrocketed instead of decreasing.

4. The Three Jev Primitives: Choice, Score, and Noul

Instead of prompting with vague system instructions, Jev APIs are strictly typed around three fundamental primitives:

5. Decision Cost Comparison: Jev vs. Budget LLMs

Here is how Jev compares against industry budget LLMs when used for high-volume automated decision tasks (e.g., 100 million routing evaluations):

6. Code Example: Implementing a Jev Router in TypeScript

Calling Jev is type-safe and directly evaluates state without string parsing:

import { JevClient } from '@typesafe/jev';

const jev = new JevClient({ apiKey: process.env.TYPESAFE_API_KEY });

// Sub-100ms decision router at $0.042 / 1M tokens
const decision = await jev.choose({
  state: incomingUserTicketText,
  question: 'Which agent pipeline should handle this request?',
  choices: ['billing_refund', 'technical_bug', 'feature_request', 'spam']
});

console.log(decision.choice);      // 'technical_bug'
console.log(decision.confidence);  // 0.984