Green Coding

Specialized AI models for software development, optimized for efficiency and sustainability. These models are designed for agentic coding tasks, multi-file editing, and integration with modern AI coding assistants.

Available coding models

High-performance models for software engineering tasks, hosted on sustainable EU infrastructure.

devstral-2-123b-instruct-2512

Mistral · Dense (FP8)

Mistral's flagship agentic LLM specifically designed for software engineering tasks. Excels at multi-file editing, codebase exploration, and powering software engineering agents.

Key strengths

• Multi-file editing capabilities.
• Codebase exploration with tools.
• Powers SWE agents (Cline, Kilo Code, Claude Code).
• 256K context for large repositories.

Benchmarks

• SWE-bench Verified: 72.2%
• SWE-bench Multilingual: 61.3%
• Terminal Bench 2: 32.6%

qwen3-coder-30b-a3b-instruct

Qwen (Alibaba) · MoE (128 experts, 8 active)

Efficient Mixture-of-Experts model optimized for coding. Only 3.3B parameters active per forward pass, making it highly efficient while maintaining strong coding capabilities.

Key strengths

• Efficient MoE architecture (lower compute costs).
• Repository-scale code understanding.
• Built-in function calling / tool use.
• Supports Yarn for 1M+ context extension.

Benchmarks

• LiveCodeBench: 70.8%
• BFCL v3: 70.8%

API usage

Call these models via the GreenPT API using OpenAI-compatible endpoints.

curl -X POST https://api.greenpt.ai/v1/chat/completions \
  -H "Authorization: Bearer $GREENPT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "devstral-2-123b-instruct-2512",
    "messages": [
      {
        "role": "user",
        "content": "Write a Python function to parse JSON files"
      }
    ]
  }'

import OpenAI from 'openai';

const client = new OpenAI({
  apiKey: process.env.GREENPT_API_KEY,
  baseURL: 'https://api.greenpt.ai/v1',
});

const completion = await client.chat.completions.create({
  model: 'devstral-2-123b-instruct-2512',
  messages: [
    {
      role: 'user',
      content: 'Write a Python function to parse JSON files',
    },
  ],
});

console.log(completion.choices[0].message.content);

from openai import OpenAI

client = OpenAI(
    api_key="YOUR_GREENPT_API_KEY",
    base_url="https://api.greenpt.ai/v1",
)

completion = client.chat.completions.create(
    model="devstral-2-123b-instruct-2512",
    messages=[
        {
            "role": "user",
            "content": "Write a Python function to parse JSON files",
        }
    ],
)

print(completion.choices[0].message.content)

OpenCode.ai integration

Use GreenPT coding models with OpenCode.ai, an open-source AI coding agent supporting 75+ LLM providers.

OpenCode.ai is a terminal-based AI coding assistant that can be configured to use GreenPT's coding models. Add the following configuration to your ~/.config/opencode/opencode.json or project-level opencode.json:

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "greenpt": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "GreenPT API",
      "options": {
        "baseURL": "https://api.greenpt.ai/v1",
        "apiKey": "{env:GREENPT_API_KEY}"
      },
      "models": {
        "devstral-2-123b-instruct-2512": {
          "name": "Mistral Devstral 2 123B"
        },
        "qwen3-coder-30b-a3b-instruct": {
          "name": "Qwen3 Coder 30B"
        }
      }
    }
  },
  "model": "greenpt/devstral-2-123b-instruct-2512"
}

Setup steps

1. Install OpenCode.ai: npm install -g opencode
2. Set your API key: export GREENPT_API_KEY=your_key_here
3. Create the configuration file as shown above.
4. Launch OpenCode and use /models to select GreenPT models.

Use cases

Ideal applications for Green Coding models.

• AI code assistants: power IDE extensions and coding copilots with agentic capabilities.
• Automated code review: analyze pull requests and suggest improvements automatically.
• Repository analysis: understand large codebases with 256K+ context windows.
• Multi-file refactoring: perform complex migrations across multiple files simultaneously.
• Test generation: generate comprehensive unit and integration tests.
• Documentation: auto-generate documentation from code and comments.

Sustainability

• Sustainable infrastructure: hosted on Scaleway's green EU data centers with low PUE (1.25) and WUE (0.25).
• Efficient architecture: Qwen3 Coder uses MoE (only 3.3B params active per request), reducing compute and energy consumption.
• Data sovereignty: full GDPR compliance with EU-based processing.
• Transparency: CO2 and energy usage tracking per request (coming soon).