🧠 Advanced AI

Reasoning Models

Models that think step-by-step before responding. Ideal for math, coding, science, and complex problem-solving.

87%
Math accuracy (MATH)
92%
Coding (HumanEval)
95%
Science reasoning
5x
Complex task improvement

How Reasoning Works

Reasoning models use extended thinking to break down complex problems before producing a final answer. This chain-of-thought approach dramatically improves accuracy on tasks requiring multi-step logic.

Question: A farmer has 17 sheep. All but 9 run away. How many sheep are left?
🧠 Thinking Process
Reading carefully: "All but 9 run away"
This means 9 sheep did NOT run away
So 9 sheep remain with the farmer
The answer is NOT 17-9=8 (common mistake)
Verified: 9 sheep are left
Final Answer
9 sheep are left.

Available Reasoning Models

🚀 mythic-o1

Most powerful reasoning model. Excels at complex multi-step problems, research, and deep analysis.

Extended thinking (up to 128K)
Multi-step reasoning
Code execution
Research synthesis

⚡ mythic-o1-mini

Fast reasoning model optimized for STEM. Great for math, coding, and scientific tasks.

Quick thinking (up to 16K)
STEM optimized
80% cheaper
Lower latency

Quick Start

Python
from mythicdot import MythicDot

client = MythicDot()

# Use reasoning model for complex problems
response = client.chat.completions.create(
    model="mythic-o1",
    messages=[
        {
            "role": "user",
            "content": """Write a Python function that finds the longest 
            palindromic substring in a string using dynamic programming. 
            Explain your reasoning."""
        }
    ]
)

print(response.choices[0].message.content)

Best Use Cases

📐

Mathematics

Complex proofs, calculus, statistics, and word problems

💻

Coding Challenges

Algorithm design, debugging, and code optimization

🔬

Scientific Analysis

Research synthesis, data interpretation, hypothesis testing

📊

Data Analysis

Statistical reasoning, trend analysis, forecasting

📝

Planning

Strategy development, project planning, decision trees

🔍

Research

Multi-source synthesis, fact verification, analysis

Reasoning Parameters

Parameter Default Description
reasoning_effort medium How much to think: "low", "medium", or "high". Higher = more tokens, better accuracy.
max_completion_tokens varies Total budget for reasoning + output. Includes hidden thinking tokens.
include_reasoning false Whether to return the thinking process in the response (beta).
Adjusting Reasoning Effort
# Low effort - quick answers, simple problems
response = client.chat.completions.create(
    model="mythic-o1-mini",
    reasoning_effort="low",
    messages=[{"role": "user", "content": "What's 15 * 23?"}]
)

# High effort - complex problems, maximum accuracy
response = client.chat.completions.create(
    model="mythic-o1",
    reasoning_effort="high",
    max_completion_tokens=50000,
    messages=[{"role": "user", "content": "Prove the Riemann hypothesis..."}]
)

Best Practices

💡 Prompting Tips

  • Be direct: Skip "think step by step" - the model does this automatically
  • Provide context: Include all relevant information in the prompt
  • Skip system prompts: User messages only for reasoning models
  • Ask for verification: Request the model to double-check its work

⚠️ Limitations

Reasoning models may take 10-60+ seconds for complex problems. They're not ideal for simple Q&A, chat, or tasks requiring low latency. Use standard models (mythic-4) for those use cases.