Phase-aware routing for AI coding agents. Planning goes to Opus; test generation goes to DeepSeek. You pay for the value — not for routing trivial code formatting at Opus rates.
Each plan includes a generous monthly quota, sized for real coding-agent workloads. Opus has its own separate quota so high-cost planning calls don't drain your main budget.
Go over your quota? You're billed at the provider's public list price plus 15% — verifiable on each provider's pricing page, never inflated. Each plan has a default monthly cap so you'll never see a runaway bill; raise it in settings if you need more.
Buy a one-time token pack if you prefer predictable bills, or bring your own provider keys (BYOK) and pay only the platform fee. Your call.
All plans include phase-aware routing, all 15+ coding models, streaming, auto-failover, and the routing-decision response headers (X-CodeRouter-Phase).
| Feature | Starter | Solo | Pro | Studio | Team |
|---|---|---|---|---|---|
| Monthly price | $19 | $29 | $99 | $179 | $299 |
| Included tokens | 2M | 5M | 30M | 50M | 80M |
| Included Opus tokens | 50K | 100K | 500K | 800K | 1M |
| Rate limit (rpm) | 100 | 200 | 600 | 1,200 | 2,400 |
| API keys | 3 | 5 | 10 | 20 | 50 |
| Overage (+15% on provider list price) | ✓ | ✓ | ✓ | ✓ | ✓ |
| Monthly overage cap | $40 | $60 | $200 | $400 | $600 |
| Top-ups (pay-as-you-go) | ✓ | ✓ | ✓ | ✓ | ✓ |
| Opus top-up | — | — | ✓ | ✓ | ✓ |
| Shared team dashboards | — | — | — | ✓ | ✓ |
| Support | Priority | Priority | Dedicated |
Same as every provider — roughly 3–4 English characters, or ~0.75 words. A typical 10K-line code base as context is ~100K tokens. The response headers X-CodeRouter-Cost and X-CodeRouter-Savings show per-request counts + what Opus would have cost instead.
Those give you one API key for many models — you still pick which model to call. CodeRouter analyzes the phase of your coding request (planning / implementing / debugging / testing / documenting) and routes to the cheapest capable model per call, automatically. OpenRouter solves access; CodeRouter solves cost.
Yes. When the phase detector identifies a planning or complex debug step, the router upgrades to Opus 4.8 or GPT-5.5 — that's why Opus quota exists. For the 80% of your calls that are incremental implementation, test writing, or doc updates, we route to Sonnet 4.6 / DeepSeek V4 Pro / Kimi / GLM-5.2, where quality is equal and cost is 10–100x lower.
Any OpenAI-compatible agent. Set your agent's base_url to https://www.coderouter.io/api/v1 and use your CodeRouter API key. Aider and Cursor (custom model) work out of the box. Claude Code via proxy config. See the setup guide.
Soft stop — we route your remaining calls to cheaper models and never block service. You get an email alert + dashboard warning. Raise your cap in settings any time (no new contract, no phone call).
Yes. Cancel from the dashboard or via Stripe portal. Your plan stays active through the end of your billing period. No retention gates, no salespeople.
No prompt content is persisted. We log only routing metadata — model used, token counts, phase classification, latency. The first 500 chars of each prompt are stored temporarily for classifier quality improvement, which you can opt out of in settings.
Yes, annual plans get 2 months free (e.g. Pro annual = $990/yr ≈ $82.50/mo). Toggle the billing period on checkout.
It compares the blended per-token cost of phase-aware routing against sending every request to a single premium model. Worked example at current list prices: routing a typical mixed workload (implementation, tests, docs to models like DeepSeek V4 at $0.435/$0.87 per 1M tokens; planning and hard debugging to premium models when needed) produces a blended cost well under $1 per 1M tokens, versus $5-15 per 1M input for premium-only usage — a 70-90%+ reduction before prompt caching, and more with it. Your exact number depends on usage mix, model selection, and cache hit rate; the dashboard shows your real savings per request.