DeepInfra

What it is for

DeepInfra is a pay-as-you-go inference API: pick a catalog model, call the OpenAI-compatible Chat Completions URL, and pay per token (or per GPU-hour for a dedicated box). Use it when the job is one HTTP call to a hosted model, not when you need a consumer chat subscription.

It is not open source as a product (many listed weights are). You add a card or pre-pay; there is no free unlimited tier.

What makes it unique: the cover is DeepInfra’s live Chat playground — DeepSeek V4.1 Flash selected, “What’s on your mind?” composer — not a GPU poster.

Key features

• OpenAI-compatible Chat Completions — same chat.completions client; base URL https://api.deepinfra.com/v1/openai (Quickstart).
• Tool calling — models invoke external functions so you can build an agent on the same endpoint (Tool Calling).
• Embeddings — OpenAI-compatible embeddings API for retrieval and search (Embeddings).
• Image generation — OpenAI-compatible images API for hosted diffusion / FLUX models (Image Generation).
• Private / dedicated models — deploy your own LLM or image weights on A100–B300 with autoscaling (Deploy Private Models).
• Speech recognition — transcribe audio with Whisper and other hosted STT models (Speech Recognition).

Who it's for

Best fit: developers who already have an OpenAI client and want a cheaper or open-weight endpoint with a printed row.

Poor fit: people who want a consumer chat app (ChatGPT) or a local runtime (Ollama).

Pricing

Verified 2026-10-22 (Asia/Shanghai) from deepinfra.com/pricing. Pay for what you use. A card or pre-pay is required. Concurrent requests default-cap at 200 per account.

Example Standard $ / 1M tokens (input / cached / output):

• DeepSeek-V4-Flash-0731 — $0.06 / $0.015 / $0.18
• Llama-3.3-70B-Instruct-Turbo — $0.10 / — / $0.32
• Meta-Llama-3.1-8B-Instruct-Turbo — $0.02 / — / $0.04
• bge-base-en-v1.5 embeddings — $0.005 / 1M input

GPU-hour (custom LLM, per-minute granularity, weekly invoice):

• A100 80GB — $1.49
• H100 80GB — $2.90
• H200 141GB — $3.49
• B200 180GB — $4.89
• B300 270GB — $5.89

Priority multiplies the token row by 1.5×; Flex by 0.8×. Image models (FLUX family) are billed per image on the same page.

Limitations / costs

• No card, no calls. Pre-pay or invoice is the other path.
• 200 concurrent requests unless you ask for more.
• Flex can be unavailable. Do not put production on Flex.
• This is not a hosted ChatGPT. The playground is a tester; the product is the API.

Alternatives

• Together AI — hosted open-model catalog with printed $/1M cards.
• Fireworks AI — serverless library plus on-demand GPUs.
• Groq — GroqCloud Supported Models table if you want that runtime.

Official tutorial

We searched for an official DeepInfra YouTube channel. The hits are third-party walkthroughs or an unrelated finance handle — no confirmed official product channel. Use the written Quickstart (first Chat Completions call in ~60 seconds).

Website

https://deepinfra.com

Source

Catalog row curated for AI Tools Nav. Cover is DeepInfra’s live Chat playground (DeepSeek V4.1 Flash composer). Dollars from deepinfra.com/pricing (Flash-0731 $0.06 / $0.015 / $0.18; A100 $1.49/hr). No official YouTube confirmed.

Last verified

2026-10-22 (Asia/Shanghai) editorial check.