CoderPlan is the pay-as-you-go API layer for Claude Code, Cursor, Codex CLI, and Gemini CLI. Same models, cache-optimized rates, credit that never expires — built for vibe coding, not for quota windows.
Credit never expires. Every call logs model, tokens, cache savings, and cost.
model: claude-sonnet
Input 34K · Cache read 94K · Output 4.2K
Cost $6.20 · Estimated savings $12.40
Usage credit $93.80 · Request succeeded
All major models
Claude, GPT, and Gemini behind one endpoint
Dual-protocol
Works with OpenAI- and Anthropic-compatible clients
New models, fast
Major model launches land here quickly
Credit never expires
Top up as you go; idle time costs nothing
High cache hits
Repeated context bills at cache-read prices
The same models as the official APIs, billed by real usage — no subscriptions, no usage windows. Start on free trial credit and run an actual task before you pay.
Membership vs pay-as-you-go
Same models and similar per-call cost — without the fixed monthly commitment.
Loved by developers
Real workflows, real bills.
API cache optimization
Agentic tools resend your project context and system prompts every turn. Cache hits bill that repeated context at a fraction of the price — that is where the savings come from.
Example based on common long-context tasks. Actual usage changes with model and task content.
Trackable spend
Model, tokens, cache reads, status, credit delta — logged per request. Run a small task, read the receipt, then decide which models earn your budget.
Quick start
Sign up, grab an API Key, point your tool at one endpoint — run your first real task tonight.
Sign up, grab an API Key. Trial credit covers one real task end to end.
One BASE_URL, one API Key. Claude Code, Cursor, Codex CLI, Gemini CLI — pick your weapon.
Credit burns only when tokens flow. Model, tokens, cache savings, cost — logged per request.
Have another question? Contact our support team anytime.
Yes — the same models, served through direct connections to the official providers. What makes it cheap is scale: the more traffic we serve, the more repeated context hits the cache, and cached tokens bill at a fraction of the standard rate. That is a cost structure, not a compromise. Run a real task on trial credit and diff the output against the official API yourself.
Claim trial credit, point Claude Code or Codex CLI at our endpoint, and ship something tonight.
The same models as the official APIs, billed by real usage — no subscriptions, no usage windows. Start on free trial credit and run an actual task before you pay.
Direct provider connections and volume-scale cache hits keep costs low, so prices can stay as low as ~4% of official list.
Prices sync with official pricing in real time; the console shows the actual amount charged.
| Model | Input price | Output price | Cache read | vs. official pricing | Ratio | Status |
|---|---|---|---|---|---|---|
GPT2 models | ||||||
GPT-5.6 SolOpenAI | $0.25 / 1M | $1.50 / 1M | $0.03 / 1M | Save ~95% | 0.35x | Online |
GPT-5.6 TerraOpenAI | $0.10 / 1M | $0.60 / 1M | $0.01 / 1M | Save ~96% | 0.35x | Online |
Claude2 models | ||||||
Claude Opus 5Anthropic | $0.36 / 1M | $1.79 / 1M | $0.04 / 1M | — | 0.5x | Online |
Claude Sonnet 5Anthropic | $0.14 / 1M | $0.71 / 1M | $0.01 / 1M | — | 0.5x | Online |
Gemini1 models | ||||||
Gemini 3.1 Pro PreviewGoogle | $0.26 / 1M | $1.54 / 1M | $0.03 / 1M | Save ~90% | 0.9x | Online |
Unit: USD/1M tokens
Ratio is the multiplier of the billing group serving the model: you pay the model's USD list price × the ratio. The vs-official column already factors in the top-up conversion.
Top up what you need, when you need it. Every $1 of credit spends like $1 on the official APIs, and credit never expires.
Kick the tires on your own codebase
Ship a small project end to end
A full week of heavy agent sessions
Heavy monthly usage and team collaboration
New here? Start on the free trial credit: connect your tool, run a real task, and top up when the results convince you.
Token and request estimates are based on GPT-5.6 Sol's current platform pricing and recent real Codex usage (about 93.5% cache hits); actual mileage varies with model choice and usage pattern.
Top-ups are processed by Stripe with international cards. International invoices are available; Chinese fapiao is not supported.