Objective AI
...Requests
...Spent
...Tokens
...OpenRouter LLMs
...Objective AI Models
The ultimate solution to AI hallucinations and unreliability
Introducing the first OpenAI-compliant LLM API with Confidence ID.
Build robust & reliable AI software with the power of AI-derived Confidence Scores.
Build the perfect Objective AI Model for your use case in the Studio.
Examples
empty
JSON Schema
openai/gpt-4.1-nano3ZQLc10LzVJtFS8vSTcFxPselectlogprobs4x
openai/gpt-4.1-nano3ZQLc10LzVJtFS8vSTcFxPselectlogprobs4x
deepseek/deepseek-chat-v3.11o7Odp8q5y4jtPicPc4QD1selectlogprobs3x
openai/gpt-4.1-mini2H6vs78dWH21ViATfRROSaselectlogprobs3x
google/gemini-2.5-flash-lite4FPWUwkeDDL23jDfAKkp4Mselectreasoning4x
openai/gpt-5-nano3K7J71U5P1ZaXkRIHfRViiselectreasoning3x
meta-llama/llama-4-scout3dTzI84P5PWaN7Mn3FmvQIselectlogprobs3x
meta-llama/llama-4-maverick1Ly5ZWNNrE1OHB1AyflQRaselectlogprobs3x
Easy to use with the OpenAI and Objective AI SDKs.
import OpenAI from "openai";
import { ObjectiveAI } from "objectiveai";
export type Category =
| "Job Application"
| "Inquiry"
| "Meeting Request"
| "Invoice"
| "Requires Human Review";
export async function categorizeEmail(emailContent: string): Promise<Category> {
const openai = new OpenAI({
apiKey: "<YOUR_OBJECTIVE_AI_API_KEY>",
baseURL: "https://api.objective-ai.io",
});
const request: ObjectiveAI.Chat.Completions.ChatCompletionCreateParams = {
messages: [
{
role: "user",
content: `Select the correct category for the following email:\n\n${emailContent}`,
},
],
model: "objectiveai/7GgWgWm7LLuk1LBd5UmYsh",
response_format: {
type: "json_schema",
json_schema: {
name: "email_categorization",
description: "Categorize the provided email into a predefined category.",
strict: true,
schema: {
type: "object",
properties: {
category: {
type: "string",
enum: [
"Job Application",
"Inquiry",
"Meeting Request",
"Invoice"
],
},
},
required: ["category"],
},
},
},
};
const completion = await ObjectiveAI.Chat.Completions.create(openai, request);
let winner_confidence = 0;
let winner_category = "";
for (const choice of completion.choices) {
// ignore choices generated by `select` models
if (choice.generate_id === null) continue;
// ignore choices that did not complete successfully
if (choice.finish_reason !== "stop" || choice.confidence === null) continue;
// get the winning choice
if (choice.confidence > winner_confidence) {
const parsedContent = JSON.parse(choice.content);
winner_confidence = choice.confidence;
winner_category = parsedContent.category;
}
}
if (winner_confidence < 0.8 || winner_category === "") {
return "Requires Human Review";
} else {
return winner_category as Category;
}
}
Top Objective AI Models
... |