Your prompt, rewritten for the model you run it on.
Paste it below. PromptIQ rewrites it the way your target model is best prompted — and shows what it costs before you ship.
Just want the numbers? Estimate cost across 15 models →
15 models · 6 providers
Anthropic, OpenAI, Google, DeepSeek, xAI, Mistral
130× cost spread
The same prompt can cost $0.01 or $1.30 per 1k runs — know before you pick
Live rates
Synced pricing, including scheduled changes — nothing hardcoded
How it works
01 Paste your prompt
Or start from one of 36 templates. Drop in files if the prompt depends on a schema, a diff, or a document — they count toward the estimate too.
02 Pick the model and the goal
Choose which of the 15 models you actually run this on, and what you're optimizing for: clarity, token efficiency, accuracy, structured output, creativity, or robustness.
03 Get the rewrite and the bill
The rewrite streams in beside your original, tuned to how that model family is best prompted — and the estimator shows what the prompt costs across every model before you commit.
Every rewrite runs the PromptIQ framework
Five checks, applied to every prompt — the difference between rewording and actually fixing.
Sharpen the task
Explicit action, a persona worth borrowing expertise from, and a defined output format.
Fill the context — honestly
Missing context gets a labeled {{placeholder}}, never an invented detail. You fill the gap; the model stops guessing.
Show, don't describe
Format-sensitive tasks get an example slot. Models copy patterns better than they follow descriptions.
Constrain the output
Format, length, structure, tone — stated, not implied.
The honesty valve
When accuracy matters, the rewrite tells the model to say "I don't know" instead of guessing. The single best hallucination fix.
Tuned per goal
Clarity · Token efficiency · Accuracy · Structured output · Creativity · Robustness
Priced and profiled: every model in the catalog
Each model carries live pricing anda prompting profile — how it prefers structure, how verbose it runs, how it handles long context, and how much scaffolding it needs. Profiles based on a provider's own published guidance are labeled as such; the rest are marked estimated, in the UI, where you can see them.
Anthropic
Claude Haiku 4.5 · Sonnet 5 · Opus 4.8 · Fable 5
OpenAI
GPT-4.1 nano · GPT-5.4 mini · GPT-5.6 Luna · Terra · Sol
Gemini 3.1 Flash-Lite · Gemini 3 Flash · Gemini 3.1 Pro
DeepSeek
DeepSeek V4 Flash
xAI
Grok 4.1
Mistral
Mistral Large 3
Know the bill before you pick the model
Paste any prompt — or drop a whole project's files — and compare real per-run and per-month costs across all 15 models, with the most efficient pick called out and rate-change warnings built in.
Questions
What does PromptIQ actually change about my prompt?+
It rewrites the prompt to match how your target model is best prompted — XML-style sections for Claude, system-role and markdown structure for GPT, context-first framing for Gemini, and so on. On top of that it runs five checks: sharpen the task, fill missing context with labeled placeholders, add an example slot when format matters, constrain the output explicitly, and add an "if you don't know, say so" instruction when accuracy matters. The task and intent stay exactly the same — only how it's asked changes.
Do I need an account?+
No. The optimizer and the cost estimator both work with no sign-up. Saved prompts and custom templates are stored in your browser. Accounts arrive at launch, and saved work moves with you.
Which models are supported?+
15 models across 6 providers — Anthropic, OpenAI, Google, DeepSeek, xAI, and Mistral. Every one is priced in the estimator, and every one has a prompting profile the rewrite engine uses.
How accurate are the cost estimates?+
Pricing is exact — rates are synced from a maintained benchmark database, including scheduled changes (a model whose intro pricing expires is flagged before it bites you), and any rate we haven't verified against a provider's live page is labeled as an estimate rather than presented as fact. Token counts are a cross-model approximation: we count with one tokenizer across all providers, so counts are close but not billing-grade. Good enough to choose a model; not a substitute for your provider's invoice.
How do you know how each model likes to be prompted?+
From each provider's own published prompt-engineering guidance. Where a model has a dedicated guide, its profile is marked provider-documented. Where it doesn't, we use the model family's general guidance and mark the profile estimated — visibly, in the UI. We don't invent specifics to fill gaps.
Is this only for developers?+
No. The templates cover 5 categories — Developer, Content, Business, Marketing, Data & Research — with 36 starting points, plus any custom template you save. Developer prompts are the deepest set, but a newsletter, a client proposal, or a survey analysis optimizes the same way.
What happens to my prompts?+
Your prompt is sent to the rewrite engine to be rewritten, and that's it. Saved prompts and custom templates live in your browser's local storage, not on a server. Nothing is sold, shared, or used for training.
What does it cost?+
Free while it's in preview, with no account required. Pricing at launch is metered by tokens processed — you pay for what you actually run, not a flat monthly tier. The rate isn't set yet, and we'd rather say that than post a number we'd change.