LLM observability
Chi phí & hiệu năng LLM
Cost per provider per task. Alert khi vượt budget tenant. Fallback rate < 5% là healthy.
Cost / report (avg)
$0.22
Target < $0.30
Spend tháng này
$287
▲ +12% MoMFallback rate
2.8%
Primary failed → fallback
LLM calls 30d
4,127
Spend per provider (30d)
Anthropic Claude Sonnet 4.6$184.30
OpenAI GPT-4o$79.20
NVIDIA NIM Llama 3.3 70B$23.50
Task performance
| Task | Primary | Fallback rate | p95 latency | $ / call avg |
|---|---|---|---|---|
reflection_score | Anthropic | 1.2% | 2.8s | $0.018 |
narrative_dimension | Anthropic | 3.8% | 4.1s | $0.031 |
narrative_recommendations | OpenAI | 2.1% | 3.5s | $0.024 |
validator | Anthropic (Opus) | 5.4% | 6.2s | $0.082 |
Top spenders (tenants)
| Tenant | Calls 30d | Spend | Budget | Status |
|---|---|---|---|---|
| Vietcombank | 1240 | $87.40 | $150 | 58% of budget |
| BIDV | 820 | $64.10 | $120 | 53% of budget |
| RMIT Vietnam | 1102 | $71.30 | $80 | 89% of budget |
| Samsung VN | 210 | $14.80 | $100 | 15% of budget |