Where to get API access to models: number of models, starting credit, payment methods and the lowest prices.
Updated
One account and one API key for models from Anthropic, OpenAI, Google and open-weight families, with each request routed to a provider that serves the model. Handy when you want to switch models without signing up again.
Anthropic's own API for the Claude models, billed per token and managed in the Claude Console. The direct route for apps and agents built on Claude.
Direct access to OpenAI's GPT models, along with its image, speech and embedding models, paid per token. The starting point for apps built on OpenAI technology.
Google's web studio for trying Gemini and Gemma models and creating Gemini API keys. Some models have a free tier with rate limits; paid use runs through Google Cloud billing.
Open-weight models served on Groq's own LPU chips, reachable through an OpenAI-compatible API. A free plan with rate limits makes it easy to test fast responses.
Cloud platform for open-weight models that combines a serverless inference API with fine-tuning and GPU clusters, aimed at teams running open models at scale.
Inference platform for open-weight models: a serverless API, fine-tuning and dedicated deployments for production apps, with a small credit to start.
Pay-per-token access to open-weight models hosted by DeepInfra, through an OpenAI-compatible API — a low-cost way to call open models without running servers.
Mistral AI's developer platform, formerly La Plateforme, with the API for Mistral models. A free mode works without a payment card.
DeepSeek's first-party API for its own models, using the OpenAI-compatible request format, so existing client libraries work with a changed base address.
xAI's developer API for the Grok models, for apps that want to build on Grok directly rather than through the chat app.
Open-weight models running on Cerebras wafer-scale chips, offered as an inference API built for very quick responses. New accounts get trial credit.
Runs models on Cloudflare's GPU network, called from Workers or a REST API, with a daily free allowance measured in Neurons. A natural fit for apps already on Cloudflare.
A single Hugging Face account and token to call models of the Hub through partner inference providers, with a small monthly credit for free experiments.
To use a model in your own app, script or agent you need API access: a key, a price per token and a limit on requests. This page lists the providers that sell it. Some are vendors offering their own models — Anthropic, OpenAI, Google, Mistral, DeepSeek, xAI. Others are platforms that serve many models, including open-weight ones: OpenRouter, Together AI, Fireworks, DeepInfra, Groq, Cerebras, Cloudflare Workers AI and Hugging Face.
Each card links to the provider's pricing or rate-limit page and shows the date we checked it.
An aggregator such as OpenRouter gives you one account and one key for models of many vendors, which is convenient when you want to switch or compare. A vendor's own API usually receives new models and features first and may offer discounts such as prompt caching. Many teams use both: a vendor API for their main model and an aggregator for experiments.
Free plans and limits change often, sometimes monthly. The prices behind "From" are refreshed daily from models.dev and LiteLLM, but the provider's own page is the final word. For a model-by-model view of who sells what, open any leaderboard and click a row: the details list providers from the cheapest.