AI Models & APIs

Beneath most products in this area sits a model reachable over an endpoint, and this category holds those endpoints and their surrounding machinery. It spans hosted inference for text, code, images, audio and embeddings; open-weight models you can serve on your own hardware; routing gateways that spread traffic across providers with failover; fine-tuning services; vector storage used to ground responses in your own content; evaluation suites that score outputs against a fixed test set; token and cost observability; and moderation endpoints. Interfaces converge on streaming HTTP with tool calling, structured output matching a supplied schema, and batch submission.

Application developers, machine learning engineers and data teams use AI Models & APIs to add classification, summarization or extraction to existing products, and to build assistants over private documents. The comparison axes are context length, time to first token, rate limits, token cost in both directions, the license on open weights, data residency, whether prompts and completions are retained or used for training, and the deprecation policy.

Benchmark on your own task with your own prompts rather than published rankings. Read retention terms and any indemnity for generated output, since generated code can resemble licensed source closely enough to matter and still needs review. Pin versions and keep an abstraction layer so providers can be swapped. Expect per-token billing, provisioned throughput, free developer tiers, or the hardware cost of self-hosting.