Credit Calculator
VantagePrompt charges credits from the real upstream cost of a run, using credits = max(1, ceil(cost_usd / 0.0005)). Enter the tokens a run uses and this returns the exact figure — no estimate, the production formula.
Model prices are read live from the OpenRouter catalog, the same source the optimizer's model picker uses.
What one run costs on every model
Priced against the average production run — 3,700 input and 1,500 output tokens across its model calls, measured August to October 2026 — so the models are comparable. Computed with the same credit calculator the billing path uses, from catalog prices refreshed hourly.
| Model | Upstream cost | Credits |
|---|---|---|
| DeepSeek: DeepSeek V4 Flash 0731 | $0.000492 | 1 |
| Google: Gemini 2.5 Flash Lite | $0.000970 | 2 |
| OpenAI: GPT-6 Luna | $0.001120 | 3 |
| DeepSeek: DeepSeek V4.1 Flash | $0.001148 | 3 |
| Tencent: Hy3 | $0.001280 | 3 |
| OpenAI: GPT-4o-mini | $0.001455 | 3 |
| DeepSeek: DeepSeek V3.2 | $0.001532 | 4 |
| Qwen: Qwen3 235B A22B Instruct 2507 | $0.001718 | 4 |
| StepFun: Step 3.7 Flash | $0.002465 | 5 |
| OpenAI: GPT-5.6 Luna | $0.002540 | 6 |
| OpenAI: GPT-5.4 Nano | $0.002615 | 6 |
| Meta: Llama 4 Maverick | $0.002795 | 6 |
| MiniMax: MiniMax M3 | $0.002910 | 6 |
| Google: Gemini 3.1 Flash Lite Preview | $0.003175 | 7 |
| Xiaomi: MiMo-V2.5 | $0.004480 | 9 |
| DeepSeek: DeepSeek V4 Pro 0423 | $0.005580 | 12 |
| Google: Gemini 3 Flash Preview | $0.006350 | 13 |
| Google: Gemini 3.6 Flash | $0.008400 | 17 |
| Google: Gemini 3.8 Flash | $0.009240 | 19 |
| OpenAI: GPT-5.4 Mini | $0.009525 | 20 |
| Z.ai: GLM 5.2 | $0.0102 | 21 |
| Anthropic: Claude Haiku 4.5 | $0.0112 | 23 |
| SpaceXAI: Grok 4.7 | $0.0164 | 33 |
| Anthropic: Claude Sonnet 5 | $0.0224 | 45 |
| OpenAI: GPT-6 Sol | $0.0224 | 45 |
| MoonshotAI: Kimi K3 | $0.0247 | 50 |
| Google: Gemini 3.1 Pro Preview | $0.0254 | 51 |
| OpenAI: GPT-5.4 | $0.0318 | 64 |
| Anthropic: Claude Sonnet 4.6 | $0.0336 | 68 |
| Anthropic: Claude Opus 5.5 | $0.0448 | 90 |
| OpenAI: GPT-5.6 Sol | $0.0448 | 90 |
| Anthropic: Claude Opus 5 | $0.0560 | 112 |
| Anthropic: Claude Opus 4.8 | $0.0560 | 112 |
| Anthropic: Claude Opus 4.7 | $0.0560 | 112 |
| Anthropic: Claude Opus 4.6 | $0.0560 | 112 |
| Anthropic: Claude Fable 5.1 | $0.1120 | 224 |
| OpenAI: GPT-6 Astra | $0.1120 | 224 |
How far does that go on each plan?
Monthly allowance divided by the credits computed above. It moves with your inputs, because a run's cost does.
| Plan | Monthly credits | Runs at this cost |
|---|---|---|
| Free | 25 | 1 |
| Starter | 600 | 46 |
| Plus | 1,200 | 92 |
| Pro | 2,800 | 215 |
| Business | 2,000 | 153 |
Why the app deducts a different number first
The upfront deduction is a reservation sized by model cost class. The run reconciles against the real cost afterwards: it refunds any excess and charges any shortfall from your remaining balance.
| Cost class | Reserved upfront |
|---|---|
| lite | 1 credit |
| mid | 5 credits |
| premium | 30 credits |
Frequently asked questions
How credits are computed, and why the upfront deduction differs.
- How are credits calculated in VantagePrompt?
- Credits are charged from the real upstream cost of the run, not from a flat per-run fee. The formula is credits = max(floor, ceil(cost_usd / usd_per_credit)), where cost_usd is what the model provider charged for the tokens the run actually used. That is the same formula this calculator applies.
- Why does the app hold a different amount before my run finishes?
- The amount held upfront is a reservation sized by the model cost class, not the final charge. When the run completes, the real cost is reconciled against it: any excess is refunded to your balance and any shortfall is charged from what is left of it. This calculator shows the reconciled figure — the one you end up paying.
- Why is one optimization more than one model call?
- A full run can classify your input, select a template, generate the expansion, and score the result. On the free auto route the classifier and judge are skipped, which cuts it to a single call. Enter the combined tokens across the calls you expect if you want a whole-run figure.
- Are these model prices current?
- They are read from the live OpenRouter catalog, the same source the optimizer model picker uses, and refreshed hourly. If the catalog is unreachable the page falls back to manual price entry so the formula is still usable.
See the plans side by side.
Allowances, model access, and seat pricing.