Why does the token count differ from what my provider bills?

Two reasons. First, providers count the full request, which includes the chat template — role markers, message delimiters and any injected system content — that a bare text counter does not see; expect a small fixed overhead per message. Second, different model families use different tokenizers, so the same text genuinely produces different counts on different models. Use the counter for planning and the provider’s usage dashboard for reconciliation.

Is my prompt sent anywhere when I use these tools?

No. Tokenization runs locally in your browser using the tokenizer implementation itself, not an API call. The counter, cost calculator, context planner and prompt tools never transmit your text, which is what makes them safe for prompts containing proprietary content or customer data.

How current is the pricing and model data?

Each comparison and pricing page shows when it was last checked. Provider pricing changes often, and new model versions appear on no fixed schedule, so treat these figures as a planning baseline rather than a quote. For anything contractual, confirm against the provider’s own pricing page.

Does a bigger context window mean I should use it?

Not automatically. Cost scales with tokens actually sent, so filling a large window on every call gets expensive quickly, and retrieval quality often degrades as irrelevant content crowds the prompt — attention is finite regardless of the stated limit. A focused prompt with well-chosen retrieved passages generally beats one that pads the window because the space is available.

AI Engineering Tools — Tokens, Cost, Context & Prompts

Tooling for the practical questions of building on language models: how many tokens, how much money, which model, and why did the output change.

18 tools · Reviewed by Mimamsa, Founder & Engineer, CodeLint.Dev

Working with language models involves a set of calculations that are tedious, easy to get wrong, and expensive when you do. How many tokens is this prompt? Will it fit alongside the retrieved documents and the expected response? What does that cost at a million calls a month, and how does that change if you switch providers?

The counters and calculators here answer those directly, using the real tokenizers rather than the word-count-times-1.3 heuristic that quietly under-counts code and badly under-counts non-Latin scripts. Tokenization is not uniform: the same passage in English, Japanese and Hindi can differ by a factor of three or more in token count, which turns into a direct and often unexpected cost difference.

The reference guides cover the concepts underneath — how tokenization actually works, what temperature and top-p do to a distribution and why changing both at once makes results hard to reason about, how embeddings behave, and where agent frameworks help versus add moving parts.

A standing caveat: model names, prices and context limits change frequently, sometimes monthly. The comparison and pricing pages carry the date they were last checked. Confirm anything cost-critical against the provider’s own pricing page before you commit to a budget.

Calculators & analysis

Token Counter Count tokens for any text using the real tokenizers, with a per-token breakdown showing exactly where a passage splits.LLM Cost Calculator Estimate API spend from input and output token volumes across providers, and compare the same workload model by model.Context Window Planner Check whether a system prompt, retrieved documents, history and expected output all fit inside a model’s window.Model Comparison Compare models side by side on context length, pricing, modality and benchmark results.

Prompt engineering

Prompt Builder Compose structured prompts from role, context, task and format sections, with a live token count as you write.Prompt Diff Compare two prompt versions token by token to see exactly what changed between runs.System Prompt Analyzer Analyse a system prompt for length, contradictory instructions, ambiguity and structural problems.Regex from Description Generate a regular expression from a plain-English description, with each component explained.

Reference guides

Tokenization Guide How byte-pair encoding works, why token counts differ so much across languages, and what that costs.Sampling Parameters What temperature, top-p, top-k and repetition penalty actually do, and which combinations to avoid.Prompt Optimization Cutting token usage and latency without losing output quality — caching, compression and structure.Embeddings Guide Vector representations, similarity metrics, dimensionality and chunking strategy for retrieval.Prompt Patterns The prompting techniques that hold up in production, and the ones newer models made redundant.Agent Frameworks A comparison of agent frameworks, what each abstracts away, and when to skip the framework.Benchmark Leaderboard Current benchmark results across models, with notes on what each benchmark does and does not measure.Provider API Reference Endpoints, parameters, rate limits and error codes across the major model providers, side by side.AI Glossary Plain-language definitions for the terminology, from attention and quantization to RAG and distillation.AI Safety Guide Practical safety considerations for shipping LLM features: prompt injection, data leakage and evaluation.

Rough token arithmetic

Useful for a first estimate. Verify with the token counter before sizing a budget — the variance across content types is large:

Content	Approximate tokens	Note
English prose	~0.75 tokens per word	The figure most rules of thumb are based on
Source code	~1.5–2× prose	Punctuation, indentation and identifiers all split heavily
JSON payloads	~2× the equivalent prose	Braces, quotes and repeated key names dominate
CJK text	~1–1.5 tokens per character	Far denser than the per-word intuition suggests
Indic scripts	Up to 3–4× English	Frequently split to nearly one token per byte

In-depth guides

Long-form articles covering the standards and formulas behind these tools.

Frequently asked questions

Why does the token count differ from what my provider bills?: Two reasons. First, providers count the full request, which includes the chat template — role markers, message delimiters and any injected system content — that a bare text counter does not see; expect a small fixed overhead per message. Second, different model families use different tokenizers, so the same text genuinely produces different counts on different models. Use the counter for planning and the provider’s usage dashboard for reconciliation.
Is my prompt sent anywhere when I use these tools?: No. Tokenization runs locally in your browser using the tokenizer implementation itself, not an API call. The counter, cost calculator, context planner and prompt tools never transmit your text, which is what makes them safe for prompts containing proprietary content or customer data.
How current is the pricing and model data?: Each comparison and pricing page shows when it was last checked. Provider pricing changes often, and new model versions appear on no fixed schedule, so treat these figures as a planning baseline rather than a quote. For anything contractual, confirm against the provider’s own pricing page.
Does a bigger context window mean I should use it?: Not automatically. Cost scales with tokens actually sent, so filling a large window on every call gets expensive quickly, and retrieval quality often degrades as irrelevant content crowds the prompt — attention is finite regardless of the stated limit. A focused prompt with well-chosen retrieved passages generally beats one that pads the window because the space is available.

Other tool categories

Developer Tools The debugging tools you reach for when something almost works: a regex tester, a diff, a pattern generator, and a webhook inspector.Generators & Utilities Generators and converters for the values you need constantly and should never have to fetch from a server.Reference & Data The lookup tables you open a dozen times a week, each searchable and on one page.FinTech Tools Finance calculators that run entirely in your browser — your salary, balances and card numbers never leave the device.