Model Performance
How the models are doing
Accuracy, drift, speed and unit cost, read live from the model register.
Models tracked
5
from the model register
Average accuracy
87%
latest measurement
Drifting
2
medium or high drift
Average cost / call
$0.0154
805ms average
| Model | Version | Accuracy | Drift | Latency | Cost / call | 7d trend |
|---|---|---|---|---|---|---|
| Credit Risk Scoring | v4.2 | 94.1% | Low | 318ms | $0.004 | +0.8 |
| Fraud Detection Net | v2.7 | 91.6% | Low | 204ms | $0.002 | +0.2 |
| Wealth Robo-Advisory | v0.4 | 88.9% | Low | 742ms | $0.022 | +1.4 |
| Customer Sentiment LLM | v1.3 | 82.4% | High | 1.9s | $0.031 | -3.1 |
| Collections Outreach | v0.9 | 77.8% | Medium | 860ms | $0.018 | -0.6 |
Drift timeline
Monthly accuracy for every model over the last 12 months. Tap a model to hide it.
OctNovDecJanFebMarAprMayJunJulAugSep
Evaluation suites
86.3% average passRegression — golden set98% pass
Nightly, 12k cases · ran 2026-09-13
Robustness — adversarial91% pass
Perturbation suite v3 · ran 2026-09-13
Fairness — group parity82% pass
18 of 100 cohorts flagged · ran 2026-09-13
Hallucination — grounded QA74% pass
RAG benchmark, 2k prompts · ran 2026-09-13
Champion vs challenger
Sentiment v1.3
+6.2% Grounded rateChallenger · v2.0-rc3 · In test
Credit Risk v4.2
+1.9% AUCChallenger · v5.0-rc1 · Promote
Fraud Net v2.7
-0.4% RecallChallenger · v2.8-beta · Hold