Pricing
The core tool is free. No account required. Pay for the artifact, not the calculation.
No account. No credit card. Always free.
- ✓7 architecture presets
- ✓6 models across 4 providers (GPT-5.4, GPT-5.4 mini, Claude Sonnet 4.6, Claude Haiku 4.5, Gemini 2.5 Flash, DeepSeek V4 Flash)
- ✓Naive / Realistic / Worst Case cost columns
- ✓Retry, caching, batch, and infra overhead multipliers
- ✓1× / 3× / 10× growth scenarios
- ✓Lowest modeled cost with the numbers behind the ranking
- ✓Sensitivity tripwire — pick two models, see the retry-rate break-even
- ✓Shareable URL — recipient can adjust and reshare
- ✓Client-side tokenizer (paste your prompt, count stays local)
- ✓Assumptions panel with full multiplier breakdown
or $16/mo billed annually ($192/yr)
7 days for free — cancel anytime, keep access for the rest of the trial
- ✓Everything in Free
- ✓Extended model catalog — Mistral, Cohere, AWS Bedrock, Azure OpenAI, Together AI, Groq
- ✓Input token breakdown by layer — system prompt / RAG context / history / user input
- ✓Cost per MAU — $/user/month output
- ✓Global retry budget + worst-case cascade failure modeling
- ✓Full sensitivity sweep — break-evens across cache hit rate, input price, and volume for any model pair
- ✓PDF export — board-quality layout with assumptions and model selection
- ✓CSV export — all models, all columns, metadata
- ✓Slack share — post summary card to a channel
- ✓MCP server — query cost models from Claude Code, Cursor, or any MCP client
Need to take Pro for a spin and walk away with a single deliverable? Pay once, no subscription.
- –Full Pro access for 24 hours — except MCP server and the full sensitivity sweep
- –Extended model catalog — Mistral, Cohere, Bedrock, Azure OpenAI, Together AI, Groq
- –Token breakdown, cost per MAU, retry budget analysis
- –One PDF or CSV download included
- –One-time payment via Paddle
FAQ
Does the free tier ever go away?
No. The core cost modeling tool — all 7 presets, all multipliers, all three cost columns, and the share link — is permanently free and requires no account.
Who handles payments?
Paddle is our merchant of record. They process payments and handle VAT, GST, and sales tax globally — you’ll see Paddle on your bank statement. We never see your card details.
Can I cancel anytime?
Yes. Cancel from your account settings at any time. Your Pro access — including the MCP server — continues until the end of the current billing period, and cancelling during the free trial keeps access for the remaining trial days. See our Refund Policy for details on refunds.
What’s the difference between the Export Pass and Pro?
The Export Pass ($5, one-time) unlocks full Pro access for 24 hours and includes one PDF or CSV download — no subscription. Two exceptions require an active Pro subscription: the MCP server and the full sensitivity sweep (both are verified server-side). After the included download, other Pro features (extended models, token breakdown, cost per MAU, retry budget) remain available for the rest of the 24-hour window. Pro ($20/mo) unlocks everything on an ongoing basis with unlimited exports.
Is my prompt text or architecture model stored anywhere?
No. The cost table is computed in your browser, and if you use the tokenizer’s paste mode, your prompt text never leaves your device — tiktoken runs locally via WebAssembly. The one exception is the sensitivity sweep, which is computed on our server from numeric usage parameters only (token counts, call volumes, multipliers) — never your prompt text — and nothing is stored. See our Privacy Policy for the full picture.