Model Governance Registry
Every model, every version, every run
Full inventory of the models powering autonomous engineering. Cost, performance, and escalation behaviour — all reviewable, all comparable, all historical.
Registered models
7
Tokens · 24h
3.0B
Blended success rate
95.4%
Governance attestations
all signed
Active model registry
Production-attached models. Updates require change-management approval.
| Provider | Model | Version | Purpose | Cost | Tokens 24h | Success | Failure | Escalation |
|---|---|---|---|---|---|---|---|---|
| Anthropic | Claude Sonnet | 4.5 | Planning · code synthesis | $3 / 1M in · $15 / 1M out | 412M | 96.2% | 1.1% | 2.7% |
| Anthropic | Claude Opus | 4.1 | Critical risk review · escalations | $15 / 1M in · $75 / 1M out | 38M | 98.4% | 0.4% | 1.2% |
| OpenAI | GPT | 5.1 | Spec extraction · diff review | $5 / 1M in · $15 / 1M out | 188M | 94.1% | 1.8% | 4.1% |
| Gemini | 2.5 Pro | Long-context repo analysis | $1.25 / 1M in · $5 / 1M out | 224M | 92.7% | 2.4% | 4.9% | |
| Gemma | 3-27b | Local heuristics · pre-filter | self-hosted | 1.4B | 88.3% | 4.2% | 7.5% | |
| Prodia | Conviction | 3.2 | Mutation generation · falsification | self-hosted | 612M | 95.8% | 1.6% | 2.6% |
| Prodia | Sentinel | 1.8 | Security policy enforcement | self-hosted | 84M | 99.1% | 0.2% | 0.7% |
Lifecycle policy
How models enter, change, and exit production.
- 01OnboardingVendor due diligence · DPA in place · model card archived
- 02Shadow mode≥ 100k inferences in parallel before promotion
- 03PromotionApproval by AI governance committee · ledger entry
- 04Retirement30-day deprecation · final attestation · evidence archived