Aitools Test
Detect AI-written text, count prompt tokens, and estimate LLM API costs with your own model pricing. Zero server uploads — every analysis runs entirely inside your browser.
💰 Model Pricing (per 1M tokens)
Three-in-One Client-Side AI Text Analysis
Aitools Test bundles three essential LLM utilities into a single browser tab: statistical AI-content detection, tiktoken-based token counting with output estimation, and configurable cost projection using your own per-model pricing. No API keys, no account registration, no server-side processing — every computation stays inside your browser.
AI Text Detector: 5-Dimensional Statistical Analysis
The detector evaluates writing across five orthogonal dimensions — perplexity variance, burstiness patterns, n-gram repetition density, type-token vocabulary ratio, and structural regularity. Weighted scoring produces a composite probability from 0% (human-written) to 100% (AI-generated), with per-metric bar visualizations so you can see exactly which features drive the verdict. Designed for English prose above 300 characters, it also handles technical documentation, marketing copy, and academic writing.
Token Counter: Input Counting + Output Estimation
The tokenizer approximates the tiktoken cl100k_base algorithm with CJK-aware splitting — roughly 0.75 tokens per English word and 1.5 per CJK character. Beyond simple input counting, the engine detects questions and task-style prompts and estimates likely output tokens based on input complexity (simple, moderate, complex, or heavy). The breakdown shows input tokens, estimated output tokens, and projected total — critical for prompt sizing and staying within context window limits.
Cost Estimator: Bring Your Own Model Pricing
Instead of shipping with stale price tables, the cost estimator lets you input your own per-1M-token rates for any LLM — GPT-4.1, Claude 4 Sonnet, Gemini 2.5 Pro, DeepSeek V3, or any custom fine-tuned endpoint. Add as many models as you need; only rows with a price entered appear in results. Costs are computed as (input + estimated output) / 1M × your rate, with results sorted cheapest-first so you can instantly compare which provider delivers the best value for a given prompt workload.
Privacy-First: Zero-Server Text Processing
Every analysis — detection scoring, tokenization, and cost calculation — executes inside your browser's JavaScript sandbox. When you paste or type text, all computations run on your own device. Nothing is transmitted to a remote server, stored in a database, or logged by any third-party analytics. You can disconnect your internet after page load and every feature continues to work offline.
This zero-server design means proprietary prompts, unpublished articles, business documents, and personal writing stay completely confidential. Results are ephemeral — they disappear when you close or refresh the tab. No registration, no data retention, no tracking scripts in the analysis pipeline.
Frequently Asked Questions
Is the AI text detector reliable?
The detector provides a statistical probability score based on five independent metrics — perplexity, burstiness, repetition, vocabulary diversity, and structure. It is a directional indicator, not a conclusive verdict. For high-stakes use cases like academic integrity review, treat the score as one data point alongside context and human judgment. Accuracy is highest on English prose above 300 characters.
Is my text sent to a remote server?
No. AI detection, token counting, and cost estimation all execute inside your browser's JavaScript runtime. No text leaves your device. There is zero server-side processing, no database storage, and no third-party API calls in the analysis pipeline. Disconnect your internet after page load and every feature still works.
How are tokens counted in the browser?
We approximate the tiktoken cl100k_base algorithm directly in JavaScript. English words average ~0.75 tokens each, CJK characters ~1.5 tokens, with whitespace and punctuation splitting modeled after GPT-family tokenizers. Estimates typically fall within 5–10% of official tiktoken output — accurate enough for prompt sizing and cost forecasting.
Can I use my own model pricing?
Yes. Click "+ Add Model" below the text input, enter any model name and its per-1M-token price. Add as many models as you need — GPT-4.1, Claude 4 Sonnet, Gemini 2.5 Pro, DeepSeek, or custom endpoints. Only rows with a price entered appear in results. Remove any model with the × button. No pre-loaded defaults, no stale pricing tables.
How does output token estimation work?
The estimator detects questions, task verbs (explain, describe, analyze), and request patterns in your input. Complexity is classified as simple, moderate, complex, or heavy based on question count and text length. Each level applies a multiplier to input tokens: 1.0× for simple queries up to 3.5× for prompts with 6+ detected tasks. This output projection feeds directly into cost calculations so you see total (input + output) API costs.
What makes this tool different from ChatGPT detectors?
Unlike single-purpose AI detectors that send text to proprietary APIs, Aitools Test combines detection, token counting, and cost estimation in one interface with fully local processing. You see not just a score but the five underlying metrics that produced it. The token counter understands CJK text. The cost estimator accepts your own pricing instead of forcing stale defaults. And nothing you paste ever touches a server.
Who builds and maintains aitoolstest.com?
Aitools Test is built and maintained by the 345tool developer collective — an independent team focused on framework-free, privacy-first web utilities. The tool is free, requires no registration, and runs entirely in your browser. The project is sustained through non-intrusive advertising placed outside the core tool interface.
345tool Team
We are the 345tool Team
345tool is an independent developer collective engineering elite, pure client-side, and privacy-first web utilities to replace bloated internet tools.