← Transparency

The models behind GDPRchat

Behind each OpenAI name is not OpenAI — it is our best current European model for that task, updated every week as benchmarks move. Or address a specific European model directly. Every model runs on EU infrastructure, under EU data-protection law.

OpenAI names → European models

Drop-in aliases. Keep your OpenAI model name; we serve the European equivalent and upgrade it automatically.

gpt-4gpt-4-turbogpt-4ogpt-4.1chatgpt-4o-latesto1o1-minio3o3-mini
Mistral LargeParis, France (EU)
gpt-4o-minigpt-3.5-turbo
Mistral SmallParis, France (EU)
codecodexgpt-4-codegpt-4o-codecode-davinci-002
Qwen 3.5Paris, France (EU)

The mapping is generated from our live model registry — it changes the moment we approve a better model, with no action on your side.

Direct access to our fleet

Prefer to pin a specific model? Address any of these by its id.

Mistral Largemistral-large-latest
Mistral AI · Paris, France (EU)

Long-form answers, reasoning and tool use, with strong coverage across all 24 EU languages.

Context window: 128,000 tokensOfficial documentation
Mistral Mediummistral-medium-latest
Mistral AI · Paris, France (EU)

Balanced everyday answers, and it reads images (vision) reliably.

Context window: 128,000 tokensOfficial documentation
Mistral Smallmistral-small-latest
Mistral AI · Paris, France (EU)

Fast, low-cost utility work: short answers, titles, tagging and classification.

Context window: 128,000 tokensOfficial documentation
Codestralcodestral-latest
Mistral AI · Paris, France (EU)

Code generation and completion across many programming languages.

Context window: 256,000 tokensOfficial documentation
Mistral Medium 3.5mistral-medium-3.5-128b
Scaleway Generative APIs · Paris, France (EU)

Fast general answers with excellent tool-calling — our current answer and routing primary.

Context window: 256,000 tokensOfficial documentation
Qwen 3.5qwen3.5-397b-a17b
Scaleway Generative APIs · Paris, France (EU)

Very strong at code and mathematics; large context window.

A Chinese-origin model that censors or distorts certain political and historical topics. We keep an automatic safeguard on even when you select it directly: prompts about the subjects below are quietly served by a values-safe European model instead — so you never receive censored output.

Tiananmen Square (1989)Taiwan's sovereigntyUyghurs / XinjiangHong Kong (2019 protests)TibetFalun GongChinese internet censorshipcriticism of authoritarian governments
Context window: 262,144 tokensOfficial documentation

The choices behind this — right-sizing models for lower energy use, and the language-dependent political bias we guard against — are written up at frits.ai/research.