# Inference Providers: 3 cents per request?

**URL:** <https://discuss.huggingface.co/t/inference-providers-3-cents-per-request/142396>\
**Category:** Beginners\
**Created:** [February 22, 2025, 7:07pm UTC](https://discuss.huggingface.co/t/inference-providers-3-cents-per-request/142396 "2025-02-22T19:07:30Z")\
**Posts on this page:** 7\
**Page:** 1

<div class="post-metadata">

**Author:** ![Wexly](https://sea2.discourse-cdn.com/hellohellohello/user_avatar/discuss.huggingface.co/wexly/32/40233_2.png) [@Wexly](https://discuss.huggingface.co/u/Wexly)\
**Post date:** [February 22, 2025, 7:07pm UTC](https://discuss.huggingface.co/t/inference-providers-3-cents-per-request/142396/1 "2025-02-22T19:07:30Z")

</div>

Basically this is the question: the posts about using the [Inference Providers](https://huggingface.co/blog/inference-providers) seem to indicate that the price is the same as using the providers directly. But when I run a small sample with one word “Hello” it seems that $0.03 is deducted from my account. The provider’s pricing for this model says “$0.10 / M tokens” which is a far cry from $0.03 per a couple of tokens to send and receive “hello”.  
Also the amount does not seem to depend on the actual amount of tokens: if I attach a 1Mb image I still get charged the same 3 cents.

So the question is if this is a bug in billing or if there is a charge of $0.03 per request? The 0.03/request seems to add up pretty fast…

This seems to be the case for all the providers I tried. Here’s my code in case I’m doing something wrong:

```auto
import os
from huggingface_hub import InferenceClient
import base64

model_name= "Qwen/Qwen2-VL-7B-Instruct"

client = InferenceClient(
    model=model_name,
    #provider="fireworks-ai",
    provider="hyperbolic",
    #provider="nebius",
    api_key=os.environ['HF_TOKEN']
)

messages = [
    {
        "role": "user",
        "content": [
            {
                "type": "text",
                "text": """Hello"""
            }
        ]
    }
]

stream = client.chat.completions.create(
    #model=model_name,
    messages=messages, 
    max_tokens=500,
    #temperature=0.0,
    #stream=False
)

print([x.message.content for x in stream.choices])

```

---

<div class="post-metadata">

**Author:** ![John6666](https://sea2.discourse-cdn.com/hellohellohello/user_avatar/discuss.huggingface.co/john6666/32/27664_2.png) [@John6666](https://discuss.huggingface.co/u/John6666)\
**Post date:** [February 23, 2025, 3:00am UTC](https://discuss.huggingface.co/t/inference-providers-3-cents-per-request/142396/2 "2025-02-23T03:00:44Z")

</div>

@meganariley Question about pricing.

---

<div class="post-metadata">

**Author:** ![julien-c](https://sea2.discourse-cdn.com/hellohellohello/user_avatar/discuss.huggingface.co/julien-c/32/41937_2.png) [@julien-c](https://discuss.huggingface.co/u/julien-c)\
**Post date:** [February 28, 2025, 5:08pm UTC](https://discuss.huggingface.co/t/inference-providers-3-cents-per-request/142396/3 "2025-02-28T17:08:47Z")

</div>

Hi @wexly,

we’re not billing inference providers usage yet (it’s only free included credits), so we are using imperfect approximation heuristics for some of the providers.

We will be shipping accurate pricing and it will go-live in the next week. I’ll post here when it’s live.

---

<div class="post-metadata">

**Author:** ![Wexly](https://sea2.discourse-cdn.com/hellohellohello/user_avatar/discuss.huggingface.co/wexly/32/40233_2.png) [@Wexly](https://discuss.huggingface.co/u/Wexly)\
**Post date:** [March 4, 2025, 7:16pm UTC](https://discuss.huggingface.co/t/inference-providers-3-cents-per-request/142396/4 "2025-03-04T19:16:03Z")

</div>

I see, thank you very much, I was excited about this feature but got a bit scared when I saw the numbers 😳 … Sorry for a false alarm, looking forward to use this feature!

---

<div class="post-metadata">

**Author:** ![John6666](https://sea2.discourse-cdn.com/hellohellohello/user_avatar/discuss.huggingface.co/john6666/32/27664_2.png) [@John6666](https://discuss.huggingface.co/u/John6666)\
**Post date:** [March 12, 2025, 10:05am UTC](https://discuss.huggingface.co/t/inference-providers-3-cents-per-request/142396/5 "2025-03-12T10:05:35Z")

</div>

Updated.

> **[@julien-c on Hugging Face: "Important notice 🚨

For Inference Providers who...](https://huggingface.co/posts/julien-c/158943939527784)**
>
> We’re on a journey to advance and democratize artificial intelligence through open source and open science.

---

<div class="post-metadata">

**Author:** ![sereban4eg](https://avatars.discourse-cdn.com/v4/letter/s/3bc359/32.png) [@sereban4eg](https://discuss.huggingface.co/u/sereban4eg)\
**Post date:** [June 24, 2026, 4:47pm UTC](https://discuss.huggingface.co/t/inference-providers-3-cents-per-request/142396/6 "2026-06-24T16:47:42Z")

</div>

```auto
A "hello" costing ~$0.03 usually means the priced tokens aren't just your message — system prompt, any context, and minimums add up, and some providers bill per-request overhead. Worth logging input vs output tokens to see the split. Free calculator to sanity-check expected cost, like wiswes one. 

```

---

<div class="post-metadata">

**Author:** ![YL-Guo](https://avatars.discourse-cdn.com/v4/letter/y/4da419/32.png) [@YL-Guo](https://discuss.huggingface.co/u/YL-Guo)\
**Post date:** [July 3, 2026, 2:36pm UTC](https://discuss.huggingface.co/t/inference-providers-3-cents-per-request/142396/7 "2026-07-03T14:36:08Z")

</div>

I would debug this as a pricing-shape problem first, not only as a token-count problem. A one-word “hello” request is a bad sample because any fixed request overhead, provider minimum, image/VLM routing cost, or approximation heuristic will dominate the token price.

The checks I would run:

1. same model + same provider, text-only, no image input
2. very small max\_tokens, then a larger realistic max\_tokens
3. compare HF Inference Providers vs the provider directly for the same payload
4. log input tokens, output tokens, provider, model, and the charged amount for each call
5. repeat with a realistic prompt instead of “hello”

For production, I would avoid optimizing around cost per request alone. The more useful number is usually cost per completed workflow or accepted answer, because retries, longer context, images, and fallback calls can matter more than the visible user message size.