Model routing is becoming the AI procurement control plane
A seven-month GitHub cohort, package usage, and a live model catalog test whether model choice is becoming an operational control layer.
Supported by independent infrastructure and usage data
597 relevant repositories show that routing is no longer only a model-selection technique. Search-match intensity rose 152% from January to the partial July window; 57.1% of the detailed cohort supports multiple providers, 73.7% mentions cost controls, 67.8% resilience, and 78.6% observability. The median tracked package recorded 1,191,643 downloads in its latest reported month, but usage is highly concentrated: litellm alone recorded 646,121,618. The evidence supports an emerging procurement control plane, not yet a broad distribution of mature vendors.
Market snapshot
Comparable measurements from independent market surfaces.
Relevant repositories
Deduplicated, subject-filtered GitHub cohort.
Median stars
Across repositories with at least one star.
Median monthly downloads
7 real package snapshots; the 655,682,088 total is dominated by litellm.
Catalog model entries
58 model organizations in the live snapshot.
Search-match dynamics
Bars show additive GitHub Search API matches across four precise gateway and router queries; the right column shows unique subject-filtered repositories loaded for detailed analysis.
What exists inside the category
One item may contain more than one feature.
Multi-provider support
341 · 57.1%
Policy and quality routing
538 · 90.1%
Fallback and retry
405 · 67.8%
Cost and budget controls
440 · 73.7%
Observability
469 · 78.6%
Governance and access
375 · 62.8%
OpenAI-compatible interface
329 · 55.1%
Self-hosted deployment
361 · 60.5%
Representative projects
Cross-source validation
These publications are not part of the primary numeric cohort.