Skip to content
New GreenPT Code: glm-5.2 and kimi-k2.7-code now available GreenPT Code: glm-5.2 + kimi-k2.7-code
Create account
Close-up of a fern frond in soft light against a dark forest background.

GreenPT Code

Coding models built for agents

A suite of specialized open-weight AI models for software development, optimized for efficiency and sustainability. Agentic coding, multi-file editing, and drop-in integration with your AI coding assistant, all on renewable EU infrastructure.

Create an account 14-day free trial (opens in a new tab) Read the coding docs (opens in a new tab)

The case

Serious coding power, without the footprint

GreenPT Code models are built for real software engineering: they plan across steps, call tools reliably, and edit many files at once. They are open source, and they run on green EU data centres, so your coding agent gets frontier-class capability at a fraction of the carbon.

  • Open source, no secrets

    Every model is open-weight, from z.ai, Moonshot AI, Mistral, Qwen, and MiniMax. You know exactly what you run, and your code never feeds a proprietary black box.

  • Agentic by design

    Reliable tool and function calling, multi-step planning, and long-horizon workflows that finish the job.

  • Large contexts

    From 256K tokens up to 1M, enough to reason over entire repositories in one pass.

  • OpenAI-compatible

    One base URL and key. Every IDE, extension, or CLI that speaks the OpenAI API works unchanged.

  • European and private

    EU-based processing, full GDPR compliance, and no training on your code. Ever.

  • Measured, not claimed

    CO2 and energy usage tracked per request, on infrastructure with a PUE of 1.25 and a WUE of 0.25.

The models

Five open models, one API

High-performance open-weight models for software engineering tasks, hosted on sustainable EU infrastructure and reachable with a single key.

Available coding models

We do not train secret models. Every GreenPT Code model is open-weight, built by z.ai, Moonshot AI, Mistral, Qwen, and MiniMax, and hosted by us on sustainable EU infrastructure. glm-5.2 is our flagship coding model.

  • z.ai Flagship 1M

    glm-5.2

    Our flagship model for agentic software engineering. Built for long-horizon coding workflows with reliable tool calling, multi-step planning, and multi-file refactoring over large codebases.

    • Agentic
    • Functions
    • Reasoning
    • MoE 753B
  • Moonshot AI 256K

    kimi-k2.7-code

    Trillion-parameter Mixture-of-Experts coding model with efficient sparse activation. Strong tool use, long-horizon planning, and codebase-scale reasoning.

    • Agentic
    • Functions
    • MoE 1T
  • Mistral 256K

    devstral-2-123b-instruct-2512

    Agentic LLM designed for software engineering. Excels at multi-file editing and codebase exploration, and powers SWE agents like Cline, Kilo Code, and Claude Code.

    • Agentic
    • Functions
    • Dense 123B
  • Qwen (Alibaba) 256K

    qwen3-coder-30b-a3b-instruct

    Efficient Mixture-of-Experts coder: only 3.3B of 30.5B parameters active per request, cutting compute and energy while keeping strong coding capability.

    • Functions
    • MoE 30.5B
    • 1M extended
  • MiniMax 200K

    minimax-m2.5

    MiniMax reasoning model optimized for agentic coding and long-horizon workflows. The budget-friendly pick, with a 200K context and 128K max output.

    • Agentic
    • Reasoning
    • Long-horizon

Bring your own agent

Connect your coding agent

GreenPT Code is OpenAI-compatible, so any tool that supports a custom endpoint works. Most setups need the same three values, then you are done.

Base URL
https://api.greenpt.ai/v1
API key
GREENPT_API_KEY
Model
glm-5.2

Any other OpenAI-compatible IDE or extension (Cursor, GitHub Copilot BYOK, Continue) connects with the same three values.

Use cases

What teams build with GreenPT Code

  • AI code assistants

    Power IDE extensions and coding copilots with agentic capabilities and reliable tool use.

  • Automated code review

    Analyze pull requests and suggest improvements automatically, before a human reviewer looks.

  • Repository analysis

    Understand large codebases in one pass with context windows of 256K tokens and beyond.

  • Multi-file refactoring

    Perform complex migrations across many files simultaneously, guided by multi-step planning.

  • Test generation

    Generate comprehensive unit and integration tests that match how the code actually behaves.

  • Documentation

    Turn code and comments into accurate, readable documentation, kept in sync with the source.

Token compression

Cut what your agent emits

Less code and less prose means lower API cost, faster responses, and less energy per request. GreenPT Code pairs with our open-source agent skills, so the savings start on day one.

  • Honey

    Less code and less prose, plus dense agent-to-agent handoffs. Around 49% fewer code tokens at 98% of baseline quality.

  • Ponytail

    Minimal code, YAGNI first. Around 54% less code for the same task.

  • Caveman

    Terse prose. Around 75% fewer output tokens when the words matter less than the work.

GreenPT Code, in short

Which coding model should I pick?

Start with glm-5.2, our flagship: it is the recommended default for AI coding assistants and long agentic workflows. Pick qwen3-coder-30b-a3b-instruct when efficiency matters most, devstral-2-123b-instruct-2512 for SWE agents, kimi-k2.7-code for very large repositories, and minimax-m2.5 for agentic coding on a budget.

Are the models open source?

Yes. Every GreenPT Code model is open-weight: glm-5.2 by z.ai, Kimi K2.7 Code by Moonshot AI, Devstral 2 by Mistral, Qwen3 Coder by Alibaba, and MiniMax M2.5. We host them ourselves in the EU, so you get open models without running the GPUs.

How do I connect my coding agent?

Any tool that supports a custom OpenAI-style endpoint works. Set the base URL to https://api.greenpt.ai/v1, add your GREENPT_API_KEY, and choose a coding model id such as glm-5.2. Cursor, GitHub Copilot BYOK, Continue, and similar tools connect the same way.

Read the setup guides →
Can I cut token costs further?

Yes. Pair GreenPT Code with our open-source agent skills: Honey cuts around 49% of code tokens at 98% of baseline quality, Ponytail writes around 54% less code, and Caveman trims prose by around 75%. Less output means lower bills, faster responses, and less energy per request.

See Honey, our token-efficiency line →
What makes GreenPT Code sustainable?

The models run in Scaleway green EU data centres with a PUE of 1.25 and a WUE of 0.25, well below industry averages. Mixture-of-Experts models activate only a fraction of their parameters per request, and we report CO2 and energy usage per API call.

Where is my code processed?

Entirely inside the EU, with full GDPR compliance and data processing agreements on request. We never train models on your code or conversations.

Start coding greener

Your agent, our models, less carbon .

Start a free 14-day trial, no credit card. Point your coding agent at one OpenAI-compatible endpoint and run glm-5.2 and friends on 100% renewable energy.

Create an account 14-day free trial (opens in a new tab) Read the coding docs
  • 100% Renewable
  • EU Hosted
  • OpenAI-compatible