Claude Code Free Unlimited Setup with OpenCode Zen and Minimax M2.5

If you've been trying to run Claude Code without paying for a Pro subscription, you've probably already hit the same walls everyone else hits: Nvidia NIM blocks your country, Ollama demands hardware you don't have, and every "free" workaround eventually dies behind a rate limit or a verification wall.

This one is different. OpenCode Zen gives you free access to Minimax M2.5 — a model that benchmarks within 0.6 percentage points of Claude Opus 4.6 on SWE-Bench Verified. No GPU. No region lock. Works globally, including India.

I've been running this setup myself — find me at @arbindbuild if you run into issues or want to share what you're building with it.

---

Why the Other Options Fall Short

Nvidia NIM has a geography problem. A large chunk of the developer world simply can't create or verify an account. On top of that, traffic from India spikes heavily after 6 PM — the API slows to a crawl exactly when most people sit down to code.

Ollama is technically free but practically expensive. Even on a machine with an M5 Pro chip and 64GB RAM (roughly ₹4 lakh), fast models are too dumb for real coding tasks and smart models are too slow. Local AI coding is getting better, but it's not ready for most hardware setups yet.

---

What OpenCode Zen + Minimax M2.5 Actually Gives You

Before getting into the setup: is Minimax M2.5 good enough for real work?

On SWE-Bench Verified, the standard benchmark for coding ability, M2.5 scores 80.2% versus Claude Opus 4.6's 80.8%. That gap is noise. On multi-file projects (Multi-SWE-Bench), M2.5 actually pulls ahead at 51.3% compared to Opus 4.6's 50.3%. On BFCL multi-turn tool calling — which matters for agentic coding — M2.5 scores 76.8% versus Opus 4.6's 63.3%.

MiniMax reports internally that 80% of new code submissions within their own company are now generated by M2.5. That's a real-world signal worth taking seriously.

For context: a standard Claude Code Pro subscription runs on Sonnet most of the time. Opus burns through your message quota after four or five exchanges. M2.5 sits in the Sonnet-to-Opus range for coding tasks, which means you can take on large projects without hitting a ceiling.

One honest caveat: M2.5 was clearly optimized for coding. On reasoning tasks like AIME 2025, it scores 45% versus Opus 4.6's higher score. If you need a general-purpose reasoning model, Opus still wins. For coding work specifically, M2.5 holds its ground.

---

Mac and Linux Setup

Step 1: Get a Free API Key from OpenCode Zen

1. Go to opencode.ai and click Zen in the nav
2. Sign in or create an account — no billing required
3. Navigate to API Keys and click Create an API Key
4. Name it (e.g., Claude Code) and copy the key somewhere safe

Don't enable billing. The free tier is what you want.

Step 2: Install Claude Code

Even if you only plan to use Claude Code inside VS Code, you need the CLI installed first. The VS Code extension depends on it.

# From the official Claude Code site — copy the exact install command from there
claude --help  # verify it worked

Step 3: Install OpenCode

# From opencode.ai — copy the install command from the homepage
# After install, restart your terminal
opencode

Step 4: Connect OpenCode to Zen

Inside the opencode interface:

/connect

Select OpenCode Zen, paste your API key, then select Minimax M2.5 Free. Type which model are you? to confirm it loaded. Response time is around 2 seconds.

Step 5: Configure Claude Code CLI to Use M2.5

code ~/.claude  # opens the .claude config folder in VS Code

Create settings.json if it doesn't exist, then add:

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://opencode.ai/zen",
    "ANTHROPIC_MODEL": "minimax-m2.5-free",
    "ANTHROPIC_API_KEY": "your-zen-api-key-here",
    "ENABLE_TOOL_SEARCH": "true"
  }
}

If you already have a settings.json with existing config, add only the "env" block inside the existing outer curly braces. Don't overwrite the whole file.

Then:

claude
# Select theme
# Confirm "Do you want to use custom API config?" → Yes
# Type /model → select Minimax M2.5 Free

Step 6: Claude Code in VS Code (Mac)

1. Open VS Code, go to Extensions (Cmd+Shift+X)
2. Search for Claude and install the official Claude Code for VS Code extension (Anthropic, verified tick)
3. Open any project folder
4. Click the Claude Code sidebar icon, then New Session
5. Type /model and select Minimax M2.5

---

Windows Setup

Step 1: Install Node.js via Chocolatey

OpenCode requires Node.js. Chocolatey is the cleanest way to get it on Windows.

Open PowerShell as Administrator, then install Chocolatey from chocolatey.org/install. After that:

choco install nodejs
# Type Y when prompted
node -v  # should return something like v24

Close PowerShell and open a normal (non-admin) window for the rest.

Step 2: Install OpenCode

npm install -g opencode-ai

Step 3: Connect to OpenCode Zen

opencode
# /connect → OpenCode Zen → paste your API key → Minimax M2.5 Free

Step 4: Install Claude Code

From the Claude Code docs, find the Windows PowerShell install command and run it in a new terminal tab.

Step 5: Configure settings.json on Windows

Navigate to C:\Users\YourUsername\.claude\ in File Explorer. Open or create settings.json and add:

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://opencode.ai/zen",
    "ANTHROPIC_API_KEY": "your-zen-api-key-here"
  }
}

Step 6: Fix PATH Issues (if claude isn't recognized)

1. Search Edit the system environment variables in Windows search
2. Click Environment Variables → find Path under System variables → Edit
3. Add: C:\Users\YourUsername\AppData\Roaming\npm
4. Click OK on all dialogs, close all PowerShell windows, open a new one

claude  # should work now

Step 7: Claude Code in VS Code (Windows)

1. Install the Claude Code for VS Code extension (Ctrl+Shift+X → search Claude)
2. Click Trust Publisher when prompted
3. Open a folder, click the Claude Code sidebar icon → New Session
4. Type /model → select Minimax M2.5

---

How Long Will This Be Free?

Nobody knows. OpenCode Zen hasn't announced an end date for the Minimax M2.5 free tier. It could last a month, it could last a year. Use it while it's available, and don't build critical production workflows around it without a fallback.

The pricing gap is real regardless. Minimax M2.5 costs about 1/30th of Claude Opus 4.6 per input token when you're paying for API access. Even if the free tier disappears, the paid tier is cheap enough to be worth knowing about.

If this setup breaks at some point — model gets pulled, URL changes, rate limits tighten — check the OpenCode docs or reach out to me directly at @arbindbuild. I'll keep this updated as things change.

— Arbind Singh