How do you use Ox Alpha with Claude Code?

For the past few days, info about Ox Alpha has been circulating on Twitter, and nobody knows where it came from. After some digging, users concluded that the model behind Ox Alpha comes from Zhipu. I tried Zhipu (z.ai) thanks to their offer of 3M free tokens for the first 5 days, and I find it quite effective, but slow. I think Zhipu is desperate to attract more users because they offered 100M tokens to new users (I didn’t get them, since I was late).

Ox Alpha is even slower, but at least it’s free.

To do this, you need to sign up on Open Router, and generate an API Key. You don’t even need to enter your credit card. And I limited usage to $1, you never know. And disabled AI training in the settings.

Next, open the command terminal and enter this:

$env:CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS="1"
$env:ANTHROPIC_BASE_URL="https://openrouter.ai/api"
$env:ANTHROPIC_AUTH_TOKEN="API_KEY_HERE"
$env:ANTHROPIC_MODEL="stealth/ox-alpha[1m]"
$env:ANTHROPIC_DEFAULT_SONNET_MODEL="stealth/ox-alpha"
$env:ANTHROPIC_DEFAULT_OPUS_MODEL="stealth/ox-alpha"
$env:ANTHROPIC_DEFAULT_HAIKU_MODEL="stealth/ox-alpha"
$env:ANTHROPIC_SMALL_FAST_MODEL="stealth/ox-alpha"
$env:CLAUDE_CODE_SUBAGENT_MODEL="stealth/ox-alpha"
$env:CLAUDE_CODE_EFFORT_LEVEL="max"

The “API_KEY_HERE” spot is where you need to add your Open Router API key.

Then launch Claude in the same command terminal and you can use the Claude Code harness with the Ox Alpha model.

Ox Alpha allows 1M of context. It’s too good to be true and yet it really works! 265M tokens to date, for $0!

But free comes at a price: it’s slow! That’s why I launch it on tasks that absolutely don’t need my input and I launch multiple subagents so it runs at the same time.

Keep an eye on Ox Alpha’s pricing, it’s announced to be free this week but it could become paid at any moment so check. In any case, it’s the 2nd most used model on Open Router these past few days.

Since we don’t know who’s behind Ox Alpha, don’t run it on just any project.