Scaleway joins Hugging Face: which AI models are available and at what price?

The arrival of Scaleway as an inference provider on Hugging Face marks an important step for the artificial intelligence ecosystem in Europe. This partnership aims to offer a sovereign AI alternative to American giants like Together.AI or Fireworks AI, while guaranteeing an infrastructure compliant with GDPR and the AI Act.

Unlike the very broad offers of its competitors, Scaleway’s inference offer focuses on a limited number of open source AI models. The most powerful models currently available include:

• gpt-oss-120b
• qwen3-235b-a22b-instruct-2507
• llama-3.3-70b-instruct
• gemma-3-27b-it
• deepseek-r1-distill-llama-70b

These models rank among the most advanced in the open weight ecosystem, even if their diversity remains lower than what is offered by Together.AI or Fireworks AI.

Competitive pricing for AI inference

One of Scaleway’s strongest points lies in its transparent and competitive pricing. Billing is calculated per million tokens, with the first tier including 1 million tokens for free. A few examples highlight this strategy:

• deepseek-r1-distill-llama-70b: €0.90 per million tokens for both input and output.
  For comparison, Together.AI charges around $2 for the same model and token volume.
• gpt-oss-120b: €0.15 for input and €0.60 for output, a price very close to competitors.
• llama-3.1-70b-instruct and llama-3.3-70b-instruct: €0.90 for input and output, aligned with Scaleway’s highest pricing tier.

These figures show that Scaleway positions itself as an affordable choice for European developers, especially on heavy models like DeepSeek where the price gap with Together.AI is significant.

A cloud offer focused on sovereignty and European infrastructure

Beyond pricing, Scaleway emphasizes its status as a sovereign European cloud provider. Computations are carried out on GPUs hosted in European datacenters located in Paris, with a strong promise of strict GDPR compliance.

This positioning addresses the growing concerns of companies and public institutions looking to avoid exclusive dependence on American infrastructures. As Hugging Face explains in its official blog post, the integration of Scaleway allows developers to run AI models directly from the Hub interface or via Python and JavaScript SDKs with full transparency. Thanks to Hugging Face’s unified inference interface, it is possible to switch providers very easily. The benefit is clear, no longer being dependent on a single provider.

Comparison with other European AI initiatives

Scaleway is not the only player moving forward in the field of European sovereign AI. Another example is Public AI, also available through Hugging Face, which relies on a different approach: making some AI models free, but with limited usage.

However, the main drawback is that these initiatives remain scattered. A truly unified European AI platform could become a credible alternative to American giants. Today, efforts are fragmented, limiting the potential impact of a coherent and competitive sovereign offer.

Key takeaways

• As a European inference provider on Hugging Face, Scaleway is the main player officially integrated since September 2025 as the first European sovereign cloud provider on the Hugging Face Hub.
• Public AI is also an inference provider integrated into Hugging Face since September 2025. Its infrastructure is distributed with a strong European orientation, but some resources may still be served globally depending on partner clusters.
• The offer covers significant open source AI models, but in smaller numbers compared to competitors.
• Pricing is competitive, especially for models like deepseek-r1-distill-llama-70b.
• The core promise remains European AI sovereignty, with all processing operated on servers located in France.
• The European AI ecosystem would benefit from better coordination between actors like Scaleway and Public AI.