LLMs

o1 API

o1 is an AI model series specifically trained for complex reasoning scenarios involving extended internal chains of thought. It supports detailed multi-step reasoning workflows, comprehensive logical inference, extended problem decomposition, rigorous analytical exploration, and long-context reasoning tasks.

1RPC.ai1RPC.ai
Reasoning
Speed
$15.00/$60.00
Input/Output
200,000Context Window

o1

o1 is a reflective generative pre-trained transformer (GPT) released as a preview on September 12, 2024, with the full version made available to ChatGPT users on December 5, 2024. Known internally as "Q*" and later "Strawberry," o1 is capable of generating long reasoning chains before responding, improving performance on challenging benchmarks and complex problem-solving beyond GPT-4o.

The model was initially accessible to ChatGPT Plus and Team subscribers and integrated into services like Microsoft Copilot. While it requires more compute and longer response times than prior models, o1 delivers markedly superior reasoning accuracy and safety adherence.

What it’s optimized for

o1 is purpose-built for:

  • Deep, multi-step chain-of-thought reasoning in STEM fields including advanced math, physics, biology, and chemistry

  • Complex programming tasks, algorithmic problem solving, and competitive coding

  • Detailed scientific problem solving and research assistance

  • High-accuracy responses requiring careful deliberation prior to output

  • Safety-aligned outputs with adherence to defined prompt instructions and policies

Typical use cases

o1 excels at:

  • PhD-level STEM question answering and academic research support

  • Solving challenging math competitions such as the American Invitational Mathematics Examination (AIME)

  • Providing high-end coding assistance and integration in programming tools like GitHub Copilot and Microsoft Copilot

  • Technical tutoring, interactive learning platforms, and educational content generation

  • Complex multi-step reasoning tasks in scientific and technical AI applications

Key characteristics

  • Generates long internal chains of thought before answering, spending more compute to improve accuracy

  • Solves 83% of AIME math problems (compared to 13% for GPT-4o) and ranks in the 89th percentile on Codeforces coding competitions

  • Requires significantly more compute and longer processing times than previous GPT models

  • Better adherence to prompt safety rules and policies, with mitigated hallucination rates

  • Supports up to 32,768 output tokens with reasoning tokens counted separately

  • No system prompt or image input support at launch; no streaming or tool use capabilities initially

  • Utilizes reinforcement learning with human feedback and a new optimization algorithm to refine chain-of-thought reasoning

Model architecture

Built on a transformer foundation, o1 uniquely incorporates a "think before speaking" approach by generating extensive internal reasoning tokens that guide its final output.

This approach balances depth of reasoning with safety and accuracy, blending reinforcement learning from human feedback and specialized data to hone its problem-solving abilities. The model architecture supports long context windows and complex multi-step reasoning ideal for scientific and technical domains.

Why choose 1RPC.ai for o1

  • Every call is directly tied to the exact model and version used, ensuring traceability and trust in your outputs

  • Execution runs inside hardware-backed enclaves, so the relay can’t access or log your request

  • Connect to multiple AI providers through a single API

  • Avoid provider lock-in with simple, pay-per-prompt pricing

  • Privacy by design with our zero-tracking infrastructure that eliminates metadata leakage and protects your activity

Summary

o1 excels in deep, reflective thinking, setting a new standard for complex problem solving in STEM and coding. It outperforms earlier GPT models in benchmark accuracy by generating long chains of thought before responding, albeit with increased compute cost and latency. Ideal for research, academic, and developer use cases that require careful reasoning, safety, and technical expertise.

Use o1 when you need the most advanced, accuracy-focused AI reasoning available, with a willingness to trade off speed and cost for superior performance.

Like this article? Share it.

Implement

Get started with an API-friendly relay

Send your first request to verified LLMs with a single code snippet.

import requests
import json
response = requests.post(
    url="https://api.1rpc.ai/v1/chat/completions",
    headers={
        "Authorization": "Bearer <1RPC_AI_API_KEY>",
        "Content-type": "application/json",
    },
    data=json.dumps({
        "model": "o1",
        "messages": [
            {
                "role": "user",
                "content": "What is the meaning of life?"
            }
        ]
    })
)
print(response.json())
Pricing

Estimate Usage Across Any AI Model

Adjust input and output size to estimate token usage and costs.

o1 Token Costs Calculator

≈ 7,500 words
≈ 75,000 words
$6.1500Total cost per million tokens
Learn about Pricing