How to use free LLMs in Cursor, Cline, OpenCode, and VS Code

Most AI coding tools accept any OpenAI-compatible API. That means you can plug free models — GLM 5.3, Kimi K3, DeepSeek V4, Qwen 3.8 — into your editor or terminal agent instead of paying per token. You need three things: a base URL, an API key, and a model ID. This guide shows where to get them and where to paste them in seven popular tools.

Step 1. Get a key and pick a model

For coding we recommend starting with one of these providers — none of them requires a card:

Model IDs change, so copy the exact ID from the model card in our catalog — for example GLM 5.3 or Kimi K3. The coding models page lists every free option ranked by benchmarks.

Cline (VS Code, JetBrains, CLI)

1. Open Cline's settings and choose API Provider.
2. Pick OpenRouter and paste your OpenRouter key, then type a :free model ID — or pick OpenAI Compatible for NVIDIA Build or UnoRouter.
3. For OpenAI Compatible, fill in the Base URL, API Key, and Model ID from the table above.

Cline plans and edits in many steps, so it spends requests fast. OpenRouter's free limit is 50 requests a day (1,000 after a one-time $10 top-up), which is enough for a short session but not for a full workday. NVIDIA Build and UnoRouter are more generous for experiments.

Kilo Code

Kilo Code has the same provider list: choose OpenRouter or "OpenAI Compatible" in the provider settings and fill in the base URL, key, and model ID. Its architect/code/debug modes can use different models — a strong free model for architecture and a faster one for edits works well.

OpenCode

OpenCode has OpenRouter built in: run /connect inside OpenCode (or opencode auth login), choose OpenRouter, paste the key, and select a free model with /models. For any other OpenAI-compatible provider, add it to opencode.json:

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "nvidia": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "NVIDIA Build",
      "options": {
        "baseURL": "https://integrate.api.nvidia.com/v1",
        "apiKey": "{env:NVIDIA_API_KEY}"
      },
      "models": {
        "z-ai/glm-5.3": { "name": "GLM 5.3" },
        "moonshotai/kimi-k3": { "name": "Kimi K3" }
      }
    }
  }
}

OpenCode Zen also offers several models for free for a limited time, such as Nemotron 3 Ultra Free — no extra key needed.

Continue

Add a model to config.yaml with the OpenAI provider and a custom apiBase:

models:
  - name: Kimi K3 (NVIDIA, free)
    provider: openai
    model: moonshotai/kimi-k3
    apiBase: https://integrate.api.nvidia.com/v1
    apiKey: ${{ secrets.NVIDIA_API_KEY }}
    roles: [chat, edit]

For autocomplete choose a small, fast model — a large reasoning model will feel slow on every keystroke.

Aider

Aider supports OpenRouter natively:

export OPENROUTER_API_KEY=sk-or-...
aider --model openrouter/z-ai/glm-5.2:free

For other providers, use --openai-api-base with --openai-api-key and prefix the model with openai/.

Zed

Zed works with your own keys and local models on the free Personal plan. Add an OpenAI-compatible provider in settings.json and enter the key in the Agent panel settings:

"language_models": {
  "openai_compatible": {
    "NVIDIA Build": {
      "api_url": "https://integrate.api.nvidia.com/v1",
      "available_models": [
        { "name": "z-ai/glm-5.3", "display_name": "GLM 5.3", "max_tokens": 128000 }
      ]
    }
  }
}

Cursor and VS Code + Copilot

In Cursor, open Settings → Models, enter an API key, and enable Override OpenAI Base URL with the provider's base URL. Your own key applies to chat; Tab completion and some agent features still use Cursor's own models and plan limits.

VS Code with GitHub Copilot lets you add your own models in Manage Models: OpenRouter and local Ollama models work in chat even without a paid Copilot plan.

Tips that save quota

• Split roles. Use the strongest free model only for planning and hard bugs; let a fast model handle small edits.
• Keep two providers configured. When one returns 429, switch to the other instead of waiting.
• Trim context. Exclude node_modules, build output, and lockfiles — free routes often cap context below the model's maximum.
• Go local for autocomplete. A small model in Ollama or LM Studio costs nothing and has no rate limit.

More tools that support free models are in our IDE extensions and terminal agents lists.