Model Settings

Claira gives you control over model behavior through Temperature, Top P, and Reasoning level.

Default settings (new cases)

The default values for new cases are:

• Temperature: 0.1
• Top P: 0.2
• Reasoning level: none

Why these defaults:

• Temperature 0.1 favors stable, repeatable outputs for legal review.
• Top P 0.2 allows limited variation without making responses too loose.
• Reasoning none prioritizes speed for most day-to-day workflows.

What each setting does

Temperature

Temperature controls response randomness.

• Lower values (0.0 to 0.3): more consistent and focused.
• Higher values (0.7 to 1.0): more varied and creative.

For most review tasks, stay low unless you intentionally want broader exploration.

Top P

Top P controls how broad the model's token choices are.

• Lower values: tighter, more deterministic phrasing.
• Higher values: more diversity and variation.

Top P and Temperature both influence variability. In practice, adjust one at a time when tuning quality.

Reasoning level

Reasoning level controls how much extra reasoning effort the model uses before responding.

• none: fastest
• low: light extra reasoning
• medium: stronger reasoning for complex edge cases
• high: deepest reasoning, typically slowest

Task-based recommendations

• High-volume relevance coding: temperature 0.1, topP 0.2, reasoning none
• Privilege edge cases / nuanced legal distinctions: temperature 0.1, topP 0.2, reasoning low or medium
• Complex fact synthesis from difficult documents: temperature 0.1, topP 0.2-0.4, reasoning medium

Notes

• Some controls may be unavailable for models that do not support the same advanced settings.
• When tuning, change one setting at a time and validate against a benchmark set.

---

Need help? Contact us at support@claira.to.