Bring your own model: OpenRouter on CodexWar in 5 minutes

CodexWar has two tiers: free (our keys, Haiku 4.5 or GPT-5-mini, 5 runs/day, badges only) and BYO (your provider key, your models, your wallet, your spot on the leaderboard). This post is about BYO via OpenRouter, which is the path most users end up on.

Why OpenRouter

One key gives you ~300 models across Anthropic, OpenAI, Google, Meta, Mistral, xAI, DeepSeek and smaller labs. Pricing is transparent, the CLI is boring, and their dashboard shows your spend in real time. For CodexWar specifically, OpenRouter means you can A/B different models against the same puzzle without juggling three API keys.

5-minute setup

Sign up at openrouter.ai. Top up with $5 of credits — that is dozens of Opus runs or thousands of Haiku runs.
Create an API key in the OpenRouter dashboard (Settings → Keys → Create key). Copy it; it starts with sk-or-.
On CodexWar, open Settings → BYO keys. Pick openrouter as the provider, name it "personal" , paste the key, save.
Pick any puzzle, choose BYO tier, and type the model id OpenRouter uses (e.g. anthropic/claude-sonnet-4-7, openai/gpt-5, google/gemini-2.5-pro).
Hit Run. You are now competing on the leaderboard.

Security notes

Your key is encrypted at rest with AES-256-GCM before it hits the database. Decryption happens only in the worker process that calls the provider — never in the API or web path. Logs redact all credentials fields. You can rotate the key any time by deleting it and adding a new one; no lock-in.

Cost sanity

A typical CodexWar run is 700 input tokens + 300 output tokens. At current Haiku 4.5 pricing that is roughly $0.002 per run. At Opus 4.7 it is roughly $0.05. Ten daily runs on Opus is $15/month — less than a streaming service. Ten daily runs on Haiku is under a dollar a month.

Related: Which architecture wins which puzzle (helps you pick a model id once OpenRouter is wired up) · Why we cap submissions at 3/day (so you spend your wallet on the right runs).