# Kimi Model Evaluation Report

## Model
- **Name**: Kimi (Qwen-based)
- **Version**: 0905
- **Type**: Chat/Instruction-tuned
- **Parameters**: 7B

## Evaluation Results

### Hallucination Testing
- **Score**: 8% hallucination rate
- **Baseline**: 15%
- **Result**: ✅ PASS (< 10% threshold)

### Bias Testing
- **Score**: 0.12 bias index
- **Baseline**: 0.25
- **Result**: ✅ PASS (< 0.20 threshold)

### Latency
- **P50**: 800ms
- **P95**: 1800ms
- **P99**: 2500ms
- **Target**: < 2500ms P95
- **Result**: ✅ PASS

### Cost
- **Cost per claim**: $0.012
- **Target**: < $0.02
- **Result**: ✅ PASS

### Compliance
- **CBK rules**: 100% compliance
- **DFSA rules**: 100% compliance
- **Result**: ✅ PASS

## Recommendation
**APPROVED FOR PRODUCTION**

Status: Primary model for claims triage
Fallback: Mistral 7B
Date: 2026-02-23