DeepInfra API key setup

Get your key

1. Open the DeepInfra API keys page: deepinfra.com/dash/api_keys
2. Create an account and generate an API key.
3. Paste it into the Dracon BYOK provider keys, stored encrypted (AES-256-GCM), used browser-direct, works in every Dracon tool.

Get the key and test it

1. Open the DeepInfra API keys page: deepinfra.com/dash/api_keys Create a key and copy the secret straight away.
2. Point your client at this provider's base URL: https://api.deepinfra.com/v1/openai It speaks the OpenAI-compatible shape, so an OpenAI SDK client works against it with no adapter: only the base URL and the key change.
3. Prove the endpoint answers before you trust it. Send one call with no Authorization header: a live endpoint answers 401 Unauthorized, because it is waiting for a bearer key. The wording of that error differs by provider (some say invalid_api_key, others incorrect_api_key or Token is invalid), so read the 401 as the signal, not the error text.
4. Paste the key into the Dracon BYOK provider keys ; stored encrypted, used browser-direct, no relay.
5. Compare other provider plans when this subscription does not match your usage or budget.

Test the key

Both examples read the key from the DEEPINFRA_API_KEY environment variable, so the key stays out of the command line.

export DEEPINFRA_API_KEY="<YOUR_API_KEY>"

curl https://api.deepinfra.com/v1/openai/chat/completions \
  -H "Authorization: Bearer $DEEPINFRA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"meta-llama/Meta-Llama-3.3-70B-Instruct","messages":[{"role":"user","content":"hi"}]}'

import os
from openai import OpenAI

# export DEEPINFRA_API_KEY="<YOUR_API_KEY>" first
client = OpenAI(
    api_key=os.environ["DEEPINFRA_API_KEY"],
    base_url="https://api.deepinfra.com/v1/openai",
)
resp = client.chat.completions.create(
    model="meta-llama/Meta-Llama-3.3-70B-Instruct",
    messages=[{"role": "user", "content": "hi"}],
)
print(resp.choices[0].message.content)

Free tier

No recurring free tier advertised as of the last check (2026-10-27). Usage is billed per the provider's listed pricing. You pay them directly, no Dracon markup.

Rate limits: 200 concurrent requests

Paid plans compared

Every tier DeepInfra publishes, with the price and the limit as the vendor states them, last checked 2026-10-27. Confirm with the provider before you buy. You pay them directly, no Dracon markup.

Official plan source: deepinfra.com/pricing

Connection details

Base URL

https://api.deepinfra.com/v1/openai

API shape

OpenAI-compatible

Env var

DEEPINFRA_API_KEY

Models

meta-llama/Meta-Llama-3.3-70B-Instruct, Qwen/Qwen2.5-72B-Instruct

Model lineup

Llama 3.1 70B, Llama 3.3 70B, Qwen 2.5 72B, Mistral 8x7B, Mixtral 8x22B

Our verdicts on this kind of provider

Independent comparisons we keep current as pricing and free tiers shift. A guide is setup; a verdict is whether it's worth setting up.

• 22 of the 45 AI providers have a free tier. Only 7 are actually useful.
• Free AI rate limits, compared: which free tiers can do real work
• The AI price war: who cut what in July 2026