Credits

Prompt2CAD bills in credits. One credit is one cent of "list price." Everything else is just arithmetic.

Costs and usage examples

Action	Cost
New CAD model	41–140 credits in the example below
Model update	8–28 credits in the example below
Render (photorealistic image)	50 credits (fixed)
Generated inspiration image	4 credits each (fixed; the agent may request 1–4)
Parameter slider drag	0
PNG screenshot	0
Project switching, viewport rotation, anything UI	0

AI chat is charged by actual token usage. These are illustrative examples, not measured averages, fixed prices, or guaranteed completed models. The new-model example uses 10 calls, each with 100,000 input and 3,000 output tokens; the update uses 2 calls, each with 100,000 input and 4,000 output tokens. Output includes reasoning. The ranges compare 90% input cache hits with no cache hits, except the first new-model call is always uncached. Each call rounds up to whole credits, with a one-credit minimum. Other paid AI tools are excluded.

Actual usage can fall outside these ranges. A short model update usually costs less than creating a complex model from scratch. Long prompts, reference images, large conversation histories, and longer agent responses can increase the cost. Render and generated-inspiration prices are fixed; uploaded reference images are part of chat token usage instead.

The free trial

New accounts can receive up to 160 trial credits on first sign-in. AI chat is charged by actual token usage, so the exact number of actions varies with the complexity and length of the conversation, including provider-reported cache hits. The examples above illustrate how far credits can go; a render costs 50 credits.

The trial is intended to let you create a first CAD model and test a conversational or parametric update before topping up.

Trial credits are general-purpose: chat, renders, and generated inspiration images and API generation all debit the same balance.

Developer API access

Create an API key in Developer settings, also available under Developer API in the app's avatar menu. API generation and refinement use the same account credits and token-based billing as the app; there is no separate API subscription or balance.

API CAD downloads are available during the trial without a purchase. Uploading reference images and creating exports have no additional credit charge in this release; reference images used in generation contribute to the model's token usage. API artifacts are downloadable for seven days after export creation. Web-app CAD and project-file downloads retain the purchase requirement described below.

See the Developer API quickstart for authentication, generation, refinement, supported formats and usage endpoints.

When you "have to pay"

Two gates:

1. Available balance. Chat can start with any positive available balance. A call temporarily reserves up to 20 credits, or the positive balance if smaller. Conversation compaction uses a separate hold of up to 5 credits, which is settled and released before the main call starts. These are estimates, not fixed fees or maximum-response purchases. Another running request can temporarily reserve funds; wait for it to finish or top up to continue.
2. Web-app CAD / 3D and project-file exports. STEP, DXF, SKP, GLB, DAE, OBJ, STL, 3MF, and .p2cad project downloads in the app are gated until a verified top-up succeeds. The paid entitlement remains on your account forever. A balance within the chat grace margin described below does not suspend paid downloads; debt below −10 credits must be reduced before downloading. PNG export is always free. API downloads follow the trial-access rules above.

The two gates are otherwise independent. You can have a healthy balance (1,000+ credits) and still hit the export gate if you've never topped up — those credits came from the trial.

What affects a model update's cost

Credits are based on the tokens processed during the chat turn. Cost generally increases when:

• The current project has a long conversation history.
• The request includes detailed requirements or reference images.
• The agent needs several tool steps to inspect and update the geometry.
• The requested change requires rebuilding a large portion of the model.

“Model update” is a customer-facing estimate, not a separate fixed billing unit. A small dimensional change may cost less than the example; a structural rewrite may cost more. Reusing context can reduce cost when the provider reports cache hits.

Failed turns

Failed AI chat calls are not charged, including provider errors, rate limits, and timeouts. If usage cannot be verified, the temporary reservation is released rather than charged at its maximum. An abandoned reservation may remain pending until its 10-minute idle expiry is reconciled when you next access your balance. Active calls renew that lease automatically; unused funds are returned before a normal completion or error is delivered.

A generation can contain several successful calls before a later one fails. Those earlier calls, including planning or model updates, and completed auxiliary work such as conversation compaction remain charged. This is not a guarantee that an entire design is free unless it completes.

If three consecutive geometry updates fail without progress, the agent stops before another AI call to save credits. Read-only inspection does not reset that limit. Your existing CAD is retained; send a simpler or different instruction to try again. Earlier successful AI calls remain charged.

If you suspect you were charged for a failed turn, email support with the project name and approximate time. Usage records carry timestamps and can be cross-checked.

Checking your balance

• Header chip — top right of /app. Focus or hover the balance to see available, temporarily reserved, and used credits. A reservation is not an extra charge.
• Credit runway banner — appears in the chat when your balance is close to running out, using the typical new-CAD-model estimate as a guide.
• Per-action confirmation — Render explicitly confirms "50 credits" before charging.

Cost transparency

The balance is shown in credits and each completed chat turn is charged from its actual token usage. The typical new-model and model-update figures are value guides for comparing top-up amounts, not quotas or guaranteed action counts. The temporary pre-call reservation can be higher than the final charge; it is a safety hold, not an additional fee.

Small settlement grace margin

Successful calls are settled from actual token usage, which can exceed the estimate reserved at the start. An already-running call may use a grace margin of up to 10 credits below zero, shared across your account's concurrent chat calls. This is not a separate credit grant or an additional fee. New AI calls require a positive available balance again, including automatic continuations. Top up if the settled balance is zero or negative. The grace margin does not guarantee that a whole design will finish.

DeepSeek V4.1 Flash 1017 is the default CAD agent for all countries and account balances. Its customer rates per million tokens are $1.20 input, $0.024 cached input, and $4.80 output (including reasoning). Cached input pricing applies only to cache hits reported by the provider, not automatically to every refinement. Other models have their own rates. These rates apply in both the app and the Developer API, where it can be selected per generation. Total design cost depends on actual usage and auxiliary tools.

The response allowance depends on the model: DeepSeek V4.1 Flash allows up to 36,000 output tokens per call, including reasoning, while other chat models allow up to 24,000. This can increase the temporary reservation for DeepSeek; it does not make every call cost the maximum or guarantee a completed model.

At low balances, chat automatically reduces its response allowance, with a 4,096-token minimum, and can reduce thinking effort for shorter calls. It does not require you to fund a maximum-size response in advance. The account still needs positive available credits to start, and the design may require more than one call to finish. This adjustment has no separate fee.

While an insufficient-balance notice is shown, automatic refinement suggestion pills are hidden so they do not repeatedly submit blocked calls. Use the notice's top-up action, then send your next edit after your balance updates. Your existing model and exports remain subject to the usual download entitlement rules.

Why we picked this model

Usage-based billing means a small update usually costs less than a large build. Credits keep those varying costs legible in one balance, while the typical-cost estimates help you choose a top-up amount without treating every request as if it consumes the same amount of work.

Next: Top-up — how to actually pay · Developer API.