What is BYOK and Why It Saves You Money

What does BYOK mean?

BYOK stands for Bring Your Own Key. Instead of paying a single subscription that bundles AI model access, you provide your own Anthropic API key to Overclock. You pay Anthropic directly at their published rates — and Overclock adds zero markup to your API costs.

Your Overclock subscription (€12/month) covers the IDE, the desktop app, the usage dashboard, session history, and all platform features. The AI model costs are between you and Anthropic.

How bundled pricing works elsewhere

Most AI coding IDEs use an all-inclusive model. You pay one monthly fee — say $20/month for Cursor Pro or $20/month for Windsurf Pro — and get access to AI features. Sounds simple, but there are trade-offs:

• Model limits: Cursor Pro includes ~4 Claude Opus 4.6 sessions per month ($5 each from a $20 credit pool). After that, each additional session costs $5.
• Daily caps: Windsurf Pro (updated March 2026) uses a quota system with daily and weekly refresh limits. Opus 4.6 costs 2× the standard quota.
• No model choice: The provider decides which model handles your request. You can't guarantee every prompt goes to the most capable model.
• Hidden costs: The subscription price includes the provider's API costs, infrastructure, and margin. You're paying for the API — you just can't see how much.

The math: BYOK vs. bundled

Let's compare a typical month:

With Overclock (BYOK):

• Subscription: €12/month (~$13)
• API costs: You pay Anthropic directly. A moderate user might spend $15–30/month.
• Total: ~$28–43/month
• Unlimited Opus access, full model control, prompt caching savings

With Cursor Pro (bundled):

• Subscription: $20/month
• API costs: Included (but with 500 fast request limit)
• Total: $20/month
• After 500 fast requests, you're on slower models

At first glance, Cursor looks cheaper. But consider what you get: with Overclock, every single prompt can use Claude Opus 4.6 — the most capable model — with no quota. You choose the model per-prompt. And prompt caching can reduce your input token costs by up to 90%, significantly lowering that API bill.

Why unlimited model access matters

With BYOK, there's no provider-side quota on expensive models. Since you're paying Anthropic directly, you can use Claude Opus 4.6 for every single prompt if you want. No "fast request" limits, no automatic downgrading to cheaper models.

This matters most when you're working on complex tasks — large refactors, architecture decisions, debugging tricky issues — where the most capable model makes a real difference.

Prompt caching makes BYOK even cheaper

Overclock automatically configures Anthropic's prompt caching feature. Static context like your tools definition, system prompt, and CLAUDE.md file are cached server-side. On repeat requests in the same session, reading cached tokens costs just 10% of the normal input price.

For a typical session where you send many prompts, this can reduce your input token costs by up to 90%. The API bill you'd naively calculate from Anthropic's pricing page is often much higher than what you actually pay.

Who benefits most from BYOK?

• Heavy users who would hit model quotas on bundled plans
• Developers who want control over which model runs each prompt
• Cost-conscious teams who want transparent, auditable AI spending
• Anyone who values predictability — you can set per-session spending caps in Overclock

Getting started

1. Sign up at overclock.win
2. Get an API key from console.anthropic.com
3. Enter your key in Overclock's settings
4. Start coding — you'll see exact costs in your usage dashboard

The free trial gives you 7 agent prompts with no credit card required, so you can try it before committing.