AI models

OpenAI

Model, embeddings, realtime, image, and agent APIs.

Agents can call chat, responses, embeddings, image, audio, realtime, and tool-oriented APIs with clear request and response schemas.

Agent access evidence

API

native

REST API covers model, embedding, realtime, image, and audio workflows.

Evidence

Browser

strong

Dashboard is useful for keys, usage, logs, evals, and project setup.

Evidence

Account creation

strong

Self-serve signup; paid usage requires billing setup.

Evidence

Pricing clarity

strong

Pricing is public by model and modality.

Evidence

Docs quality

native

API reference and guides include curl examples, SDKs, request IDs, and model-specific notes.

Evidence

Sandbox

partial

Agents can test in separate projects and enforce usage limits before production rollout.

Readiness score

Usable

Usable: 74/100. Strongest in agent readability; weakest in machine operability.

Machine operability14/25

Agent safety18.6/25

Agent readability19/20

Auth and setup12/15

Production reliability10/15

Best for

Model calls, embeddings, speech, image generation, evals, and agent backends.

Pricing

Public per-model usage pricing with separate input, output, cached, and media dimensions.

Auth

API keys, project keys, organization/project scoping, and request IDs for support.

Account creation

Self-serve signup; paid usage requires billing setup.

Browser support

Dashboard is useful for keys, usage, logs, evals, and project setup.

API

https://api.openai.com/v1

Limitations

Key limitations: Production agents need budget controls, data-handling review, prompt-injection defenses, and model-specific evals. Model behavior, costs, rate limits, data retention, and safety policies can change by model and account tier; use evals, spend caps, and data controls for sensitive workloads. API automation needs client-side throttling, idempotency where supported, pagination limits, retries, and audit logging. Store credentials outside source control and use least-privilege scopes, short-lived tokens, or restricted keys where the product supports them.