AI & Services

Hetzner Opens a Free Qwen3.8-27B API, but a Credit Card Is Required.

Qwen3.8-27B was added to Hetzner's experimental inference API. Here are the pricing, rate limits, and caveats from the official docs, plus the credit card requirement users have reported.

· Huuush

Hetzner Opens a Free Qwen3.8-27B API, but a Credit Card Is Required

Key takeaways

  1. 1Hetzner's official docs (docs.hetzner.com) state that its experimental inference API offers two models, Qwen/Qwen3.6-35B-A3B-FP8 and Qwen3.8-27B, with a 262,144-token context window
  2. 2It's free during the experimental phase, but there's no performance or availability guarantee, no backups, and requests are limited to 10 per 60 seconds
  3. 3User JustinGorya said it felt like GPT-5.6 Luna-level performance, but replies reported that a credit card is needed to register an account. Hetzner's official docs don't mention this requirement
Still-life photo of a brass key next to a glass server module

Qwen3.8-27B has been added to the experimental inference API of Hetzner, the German cloud provider. User @JustinGorya wrote at 12:01 a.m. KST on August 19, 2026 that "Hetzner now offers Qwen3.8-27B for free," saying it felt on par with GPT-5.6 Luna or better in his testing. We checked the official Hetzner documentation directly, and Qwen3.8-27B is in fact listed.

What the official docs say

Hetzner's documentation describes the service's purpose as an experiment to "give customers affordable access to open-weight models." The models currently offered are Qwen/Qwen3.6-35B-A3B-FP8 and Qwen3.8-27B, and both support a 262,144-token context window. The API is OpenAI-compatible (https://inference.hetzner.com/api/v1), so you can connect an existing SDK by changing only the base URL and API key.

"Free" comes with conditions

Pricing is "free while in the experimental phase." However, Hetzner states that it doesn't guarantee performance or availability, doesn't create backups, and that the service isn't meant for production use. Requests are limited to 10 per 60 seconds. Connecting it straight to a commercial service just because it says "free" could cause trouble.

Why a credit card?

Something not in the official docs came up in the replies. When user @ottomancoder asked, "Do you need a credit card?", JustinGorya answered, "Yes, but it's still free." This is a personal report not stated in Hetzner's official documentation, and it may be a requirement of signing up for a Hetzner Cloud account rather than a pricing policy of the API itself. To be sure, check the sign-up screen yourself. In the same reply thread, JustinGorya added that GLM-5.2 and DeepSeek V4 Flash, which Hetzner previously offered, have already been discontinued, and that Qwen3.8-27B was activated that day and is being rolled out gradually.

Frequently asked questions

Is Hetzner's inference API really free?
Hetzner's official docs say it's "free while in the experimental phase." It has said it will give advance notice by email before any move to paid pricing, and it still doesn't guarantee performance or availability.
Are other models available besides Qwen3.8-27B?
Besides Qwen3.8-27B, Hetzner's docs also list Qwen/Qwen3.6-35B-A3B-FP8. Both models support a 262,144-token context window.
Is a credit card required?
Hetzner's official docs don't state a credit card requirement. However, user JustinGorya reported in an X reply that account registration required a credit card. It may be a sign-up policy for the Hetzner account itself, so check the registration screen directly.
Is there a rate limit?
Yes. According to Hetzner's docs, it's limited to 10 requests per 60 seconds. The service isn't designed for production use.
Is the claim that it beats GPT-5.6 Luna true?
That's user JustinGorya's personal impression, not an official benchmark. In the replies to the same post, another user questioned the assessment.

Related articles