DeepInfra

What it does

DeepInfra takes text, image, audio and video inputs sent to its hosted models through an API, and returns generated text, embeddings, rerankings, images, video or speech output depending on the model. Its catalog lists more than 100 models, including large language models, vision and OCR models, embedding and reranker models, image and video generation, and speech recognition and text-to-speech.

People typically use DeepInfra by pointing an existing OpenAI-compatible client at its API base URL and selecting a model, so existing code can switch providers with a changed base URL and key. Billing is mostly per token or per unit of model execution, with listed input, cached-input and output rates for language models. DeepInfra also offers dedicated GPU instances and clusters for deploying custom models.

DeepInfra suits developers and teams that want pay-as-you-go access to many open-source and hosted models without long-term contracts. Access requires creating an account and an API key, and prepaid balance or a card is needed before use. Limitations include a stated default cap of 200 concurrent requests per account, model availability that varies by model, and dedicated GPU deployments that require contacting sales.

What it can do

• Text generation through an OpenAI-compatible chat completions API
• Embeddings and reranking models for search and retrieval workflows
• Image generation with FLUX models priced per image
• Speech-to-text and text-to-speech models, including speech priced per minute of audio
• Dedicated GPU instances and clusters for deploying custom or fine-tuned models
• Service tiers (standard, priority, flex) selectable per request

Works with

• OpenAI SDK compatibility via custom base URL

Example workflows

Drop-in LLM replacement

Illustrative example. Input: existing OpenAI SDK code and a DeepInfra API key. Steps: set base_url to https://api.deepinfra.com/v1/openai and choose a model ID. Output: chat completions returned from the chosen model with no other code changes.

Semantic search embedding batch

Illustrative example. Input: a set of documents and an embedding model such as bge-m3. Steps: send texts to the embeddings endpoint, store vectors, and optionally rerank top results with a reranker model. Output: vectors and reordered search results, billed per million input tokens.

Before you choose

• Pricing varies widely by model and is quoted per million tokens, per image, per minute of audio or per GPU-hour, so compare the model page before budgeting.
• Each account is limited to 200 concurrent requests by default; higher limits require contacting DeepInfra.
• Priority and flex tiers change price and latency, and availability differs by model.

Listed under

• Coding
• Sandboxes and infrastructure
• AI inference
• Developers
• GPU rental
• Image generation
• LLM API
• Model API

Get started

Sign up, add a card or prepay, then generate an API key from the dashboard.

1. Create a DeepInfra account and add a payment card or prepay balance, since services are unavailable without one.
2. Generate an API key from the dashboard and store it as an environment variable.
3. Install the OpenAI SDK, set base_url to https://api.deepinfra.com/v1/openai, and call a chosen model ID.

Serverless model rates vary by model and token or media unit. Dedicated GPUs are metered by the minute at published hourly rates and invoiced weekly; examples include A100 at $1.49/GPU-hour and H100 at $2.90/GPU-hour. Custom clusters have separate terms. A card or prepayment is required.

Source: deepinfra.com/pricing, checked 29 Oct 2026. Prices can change; check before you buy.

Sources for this overview

Product pages, documentation, and pricing checked 29 Oct 2026. Each source supports part of the overview; availability and prices can change.

• Machine Learning Models and Infrastructure | DeepInfra ↗deepinfra.com
• Simple Pricing | Machine Learning Infrastructure | DeepInfra ↗deepinfra.com
• What is DeepInfra - DeepInfra ↗docs.deepinfra.com

Company website

deepinfra.com: product, pricing and policy pages

YC directory mirror

Batch and company name

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

Last checked 29 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.