Token Optimization

DeepMyst’s token optimization feature reduces the number of tokens used in your LLM interactions, lowering costs while preserving the quality of responses.

​

How Token Optimization Works

DeepMyst employs advanced compression techniques that:

1. Identify Redundancies: Our system analyzes message content to identify repeated phrases, patterns, and information
2. Compress Content: Redundant content is intelligently compressed to reduce token count
3. Preserve Context: Key information is maintained to ensure response quality isn’t compromised
4. Restore Meaning: The optimization is transparent to the model, allowing it to generate high-quality responses

The result is significantly reduced token usage without sacrificing the quality of responses – saving you money while maintaining performance.

​

Benefits of Token Optimization

​

Using Token Optimization

Enabling token optimization is as simple as adding the \-optimize flag to any model name in your API request:

from openai import OpenAI

client = OpenAI(
    api_key="your-deepmyst-api-key",
    base_url="https://api.deepmyst.com/v1"
)

# Add the -optimize flag to the model name
response = client.chat.completions.create(
    model="gpt-4o-mini-optimize",  # Note the -optimize flag
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "Explain the history of artificial intelligence"}
    ]
)

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

​

Optimization Levels (Coming Soon)

DeepMyst automatically applies the appropriate level of optimization based on context, but you can also specify a particular optimization level:

# Specify optimization level (1-5, where 5 is maximum optimization)
response = client.chat.completions.create(
    model="gpt-4o-mini-optimize-3",  # Level 3 optimization
    messages=[
        {"role": "user", "content": "Summarize the key events of World War II"}
    ]
)

​

Real-World Examples

Here are some real-world examples of token optimization in action:

​

Example 1: Long-Form Content

Without Optimization:

• Message size: 6,500 tokens

• Response size: 1,200 tokens

• Total tokens: 7,700

• Cost (at $0.01/1K tokens):$

  0.077

With Optimization:

• Optimized message size: 2,200 tokens (66% reduction)
• Response size: 1,200 tokens
• Total tokens: 3,400
• Cost: $0.034
• Savings: 56%

​

Example 2: Conversation History

Without Optimization:

• 10-turn conversation: 12,000 tokens

• New response: 800 tokens

• Total tokens: 12,800

• Cost (at $0.01/1K tokens):$

  0.128

With Optimization:

• Optimized conversation: 3,600 tokens (70% reduction)
• New response: 800 tokens
• Total tokens: 4,400
• Cost: $0.044
• Savings: 65%

​

Best Practices

For maximum benefit from token optimization:

1. Use in multi-turn conversations: The benefits compound as conversation history grows
2. Apply to information-dense queries: Content with repetition and patterns benefits most
3. Balance with reasoning: Combine optimization with reasoning techniques for complex tasks
4. Monitor quality: Watch for any impact on response quality and adjust optimization level if needed
5. Preserve key instructions: Keep important system messages and instructions clear and precise

​

Optimization Analytics

The DeepMyst dashboard provides detailed analytics on your token optimization:

These analytics help you:

• Track total tokens saved
• Monitor cost reduction over time
• Compare optimization effectiveness across different models
• Identify opportunities for further optimization

​

When to Use Token Optimization

Token optimization is particularly valuable for:

• Multi-turn conversations: As context grows, so do the savings
• Long-form content processing: When analyzing or generating extensive content
• High-volume applications: Applications making many API calls
• Fixed-budget projects: When working within strict cost constraints
• Context-window-limited scenarios: When you need to fit more content within token limits

​

Implementation Tips

To maximize the effectiveness of token optimization:

1. Start with balanced optimization: Begin with the default optimization level
2. Test with your specific use cases: Different content types benefit differently
3. Monitor quality metrics: Ensure optimization doesn’t impact critical outputs
4. Combine with other DeepMyst features: Use alongside reasoning or routing for even better results
5. Apply selectively: Use optimization where it makes the most sense for your workload