Jev API Pricing: $0.042 per Million Input Tokens, $0 Output

Price data as of 2026-10-09. Prices come from the BazaarLink model list; actual BazaarLink billing follows the model page and your invoice.

Jev costs $0.042 per million input tokens and $0 per million output tokens, with a 32,000-token context window. Only input tokens are billed, so a single decision usually costs a fraction of a cent. Prices are in US dollars. Jev is not a chat model: it takes JSON state and named questions and returns boolean, choice or score answers. For features and the response format, see our Jev evaluation guide; this page covers pricing and integration.

Jev price table

What one decision and one month cost

This is plain arithmetic, not a benchmark. The real response in the evaluation guide shows that a three-question request used 394 input tokens.

• One decision: 394 x $0.042 per million is about $0.0000165.
• A thousand decisions: about $0.0165.
• One month: 100,000 decisions a day at 500 input tokens each, over 30 days, is 1.5 billion tokens, so 1,500 x $0.042 = $63.

Longer input costs more, so removing fields you do not need from the state is the most direct way to save.

Price compared with other models

One decision here means 500 input tokens, plus 20 output tokens for chat models. This compares published list prices only and says nothing about decision quality.

All figures come from the BazaarLink model list (2026-10-09); for Haiku 5.5, long-prompt rates follow Anthropic's announcement, and actual BazaarLink billing follows the model page and your invoice. The Clef models accept images while Jev takes text only, so choose by your input type and by accuracy on your own data.

How to call it with the OpenAI SDK

Jev uses an evaluation-specific request (state plus questions), not chat completions. The example below calls it with the post method of the OpenAI Python SDK and keeps the key in the environment variable BAZAARLINK_API_KEY, never in source.

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://bazaarlink.ai/api/v1",
    api_key=os.environ["BAZAARLINK_API_KEY"],
)

result = client.post(
    "/evaluations",
    cast_to=object,
    body={
        "model": "typesafe-ai/jev",
        "state": "The support agent issued a full refund and apologised for the delay.",
        "questions": {
            "refunded": {"type": "boolean", "instructions": "Was a refund issued?"}
        },
    },
)
print(result["answers"]["refunded"]["probability"], result["usage"])

The full fields for the three question types (boolean, choice, score) and a sample response are in the evaluation guide and the API Reference.

Before you rely on it

• TypeSafe, the model's maker, says English is the primary training language and where accuracy is currently best. Other languages work but not equally well, so test on your own data first (source: TypeSafe model docs).
• The limit that applies on BazaarLink is the 32,000 tokens on the model list. Questions in one request share the same state, so sending several questions together uses fewer tokens than sending them separately.
• Jev returns structured answers only and writes no explanatory text. Use a chat model when you need explanations or rewriting.

Before you set a budget

Pricing this low makes it easy to skip the planning, but volume adds up, so three habits help:

1. Read the usage field. Every response reports input and output token counts. Send a sample of 100 real requests and multiply the total input tokens by the rate above.
2. Pack questions together. Several questions about one state share the same input, so a single request with five questions costs far less than five separate requests.
3. Measure accuracy before volume. A cheaper decision is only a saving if it is right often enough for your task. Label a small sample and compare Jev with the alternative you use today.

FAQ

How much does the Jev API cost per million tokens?

According to the BazaarLink model list on 2026-10-09, $0.042 per million input tokens and $0 for output, which is about $42 per billion input tokens. Actual BazaarLink billing follows the model page and your invoice.

Is Jev output really free?

The listed output price is $0, and the response usage still reports output_tokens. Cost depends only on input tokens, so the way to control it is to keep the state and questions short.

What does one Jev decision cost?

Using the 394 input tokens in the real response shown in our evaluation article, about $0.0000165, or roughly $0.0165 per thousand. Your figure depends on the length of the state and questions you send.

What is the Jev context limit?

The BazaarLink model list shows 32,000 tokens, counted over the serialized state plus the questions.

Can I call Jev through /v1/chat/completions?

No. Jev is evaluation-only. Call /api/v1/evaluations; chat completions returns HTTP 400. The default /api/v1/models list does not show it; add ?output_modalities=decisions to query it.

Does BazaarLink add a markup on Jev?

No. It bills at the model's list price. Prices are in US dollars, and Taiwan-based customers can pay in NTD and receive a Uniform Invoice.

Related

• Jev model page: /en/models/typesafe-ai-jev
• All models and live prices: /en/models
• Features, question types and response format: Jev: Get Structured Evaluation Results from State
• Vendor source: TypeSafe model docs

TWD billing · Taiwan invoices · leading AI models · OpenAI-compatible API