Green Router
AI inference service routing to popular open-weight and open-source models on sustainable, sovereign infrastructure.
Green Router is an AI inference service that provides access to the most popular open-weight and open-source models through a single API. Requests are routed to models hosted by our green and sovereign hosting provider, Scaleway, ensuring sustainable and privacy-focused AI infrastructure.
Compatible with the OpenAI SDK
Green Router models are served through the same OpenAI-compatible endpoint
as the rest of the GreenPT API at https://api.greenpt.ai/v1. See
Get started for client setup.
Available models
Green Router exposes foundation models without a GreenPT system prompt, you supply your own system prompt and parameters. See the Models catalog for full specs and the Pricing page for token costs.
| Model | Provider | Notes |
|---|---|---|
mistral-small-3.2-24b-instruct-2506 | Mistral | Recommended general-purpose starting point. |
gpt-oss-120b | OpenAI | Large open model with vision and long-context support. |
gemma-3-27b-it | Multimodal reasoning for chat and image-aware tasks. | |
qwen3-235b-a22b-instruct-2507 | Qwen | High-context multilingual reasoning (250k window). |
qwen3-coder-30b-a3b-instruct | Qwen | Coding model for generation, completion, debugging. |
qwen3.5-397b-a17b | Qwen | State-of-the-art for code, agents, and reasoning. |
devstral-2-123b-instruct-2512 | Mistral | Agentic coding model with multi-file reasoning. |
llama-3.3-70b-instruct | Meta | Instruction-following with strong multilingual chat. |
voxtral-small-24b-2507 | Mistral | Audio-aware chat for transcription-adjacent use cases. |
List available models
Fetch the live list of Green Router models from the API.
GET /v1/modelscurl https://api.greenpt.ai/v1/models \
-H "Authorization: Bearer sk-your_api_key"Why Green Router?
- Sustainable infrastructure: models run on Scaleway's renewable-powered EU data centres (PUE 1.25, WUE 0.25).
- EU data sovereignty: full GDPR compliance with EU-based processing.
- Single API, many models: switch between providers without rewriting your integration.
- No vendor lock-in: OpenAI-compatible interface means existing clients work out of the box.