Open source models, served fast.

DeepSeek, Qwen, Gemma and Llama on our infrastructure, behind one API that speaks the OpenAI format.

api.infere.dev/v1
ResponseIdle

Time to first token< 300 ms
Open models4+
Uptime99.9%
From per 1M tokens$0.10

Open models, one catalog

Pick a model by id. Prices are per million tokens.

DeepSeek V3

deepseek-v3

128K$0.27 / $1.10
Qwen 2.5 72B

qwen-2.5-72b

128K$0.35 / $0.40
Llama 3.3 70B

llama-3.3-70b

128K$0.23 / $0.40
Gemma 3 27B

gemma-3-27b

128K$0.10 / $0.20
Median< 300 ms

Time to first token, measured across the catalog.

Uptime99.9%

Rolling 90-day availability across all regions.

Live throughput

Readings stream as the model warms up.

deepseek-v3
Tokens per second42
Sustained across a 4K-token reply.

Drop-in OpenAI compatible.

Keep your client and change the base URL.

Built to be trusted

No prompt logging

Requests are not stored or used for training.

Region pinning

Choose where your requests are served.

Uptime target

Models, prices and limits are published in the documentation.

Change the base URL, keep your code.

Point any OpenAI client at infere.dev, pick an open model by its id, and stream the reply. No new SDK.

request.sh
curl https://api.infere.dev/v1/chat/completions \  -H "Authorization: Bearer $INFERE_API_KEY" \  -H "Content-Type: application/json" \  -d '{    "model": "deepseek-v3",    "stream": true,    "messages": [{"role": "user", "content": "Explain KV caching in one line."}]  }'
response.json
{  "id": "chatcmpl_8f2a91",  "model": "deepseek-v3",  "choices": [    {      "message": {        "role": "assistant",        "content": "KV caching stores attention keys and values so each new token reuses past work."      },      "finish_reason": "stop"    }  ],  "usage": { "prompt_tokens": 14, "completion_tokens": 19 }}