Machine Learning is the largest area on the network: 1,037 providers, indexed at machine-learning.apievangelist.com. It spans the full ML lifecycle — training platforms, model serving, inference endpoints, MLOps tooling, and the applied vision and language services built on top.
It is also the highest-scoring area we track, and the reason is worth sitting with.
The standouts
| Provider | Score | Band | APIs |
|---|---|---|---|
| Amazon Kendra | 78.3 | Exemplar | 8 |
| Amazon Rekognition | 77.4 | Exemplar | 10 |
| Amazon SageMaker | 76.4 | Exemplar | 9 |
| Amazon Entity Resolution | 75.8 | Exemplar | 3 |
| GitHub Copilot | 75.2 | Exemplar | 12 |
| Anthropic | 75.1 | Exemplar | 24 |
| Databricks | 75.0 | Exemplar | 57 |
| Claude | 73.6 | Exemplar | 9 |
| IBM | 72.7 | Exemplar | 55 |
| Replicate | 70.7 | Exemplar | 16 |
| Teradata | 70.7 | Exemplar | 11 |
Eleven exemplar-band providers in the top of one area. Compare that to financial services, profiled today, where zero providers reach exemplar across 125 institutions.
What the area tells you
The scores cluster high because ML APIs were born after the conventions settled. There is no legacy SOAP surface to carry forward, no 2009 REST design to maintain compatibility with. A provider shipping an inference endpoint in the last few years started with OpenAPI, structured errors, token-based auth, and a pricing page as defaults rather than retrofits.
Note the API counts against the scores. Amazon Entity Resolution scores 75.8 on three APIs. Databricks scores 75.0 on fifty-seven. Score and surface area are close to uncorrelated here — what the Kin Score measures is completeness of description, not size of product. A tightly-specified three-API service beats a sprawling under-documented one, and should.
The Amazon concentration at the top is its own observation. Kendra, Rekognition, SageMaker, Entity Resolution, Polly, HealthImaging, and QuickSight all land in the exemplar band. That is a house style being applied consistently across an ML portfolio — the same specification discipline shipped with every service, rather than each team choosing its own standard.
The long tail
1,037 providers and roughly thirty at the top means the distribution is steep. The exemplar cluster is real, but so is the mass of inference startups shipping an endpoint and a README. The area’s average is nowhere near its ceiling.
Takeaway
Machine Learning is the best-described area on the network because it is the youngest — it inherited good defaults instead of legacy. The lesson for older sectors is not that ML teams are better engineers. It is that specification discipline is cheap when you start with it and expensive when you retrofit it.
Browse the area at machine-learning.apievangelist.com.