Token Counter

Context budgeting is guesswork without a real count, and model choice is guesswork without a real price. Paste a prompt, see tokens, then compare projected cost across the whole hosted-model catalogue rather than the four models you happen to remember.

Fetches a static price table from this site. Your text is never sent anywhere, and the request carries nothing but its own URL.

Input text

tokens: 0; ±0
characters: 0
words: 0
lines: 0

Estimate, not an exact count. No production tokenizer runs client-side, so this uses a segment-weighted character model, typically within ±10% on prose and ±15% on code. For an authoritative figure, call the provider's own counter, e.g. /v1/messages/count_tokens.

Projection

Calls

Output / input25% · 0 tok

Input from cache0%

Cost · 4 models

loading prices

Model	$/1M in	$/1M out	Cache read	Context	Est. cost
Claude Opus 4.6pinned Most capable Anthropic reasoning and agentic coding model.	$5.00	$25.00	—	1M	$0.00
Gemini 2.5 Flashpinned High-speed, multimodal Google model with 1M context window.	$0.300	$2.50	—	1M	$0.00
GPT-4opinned OpenAI flagship versatile multimodal model.	$2.50	$10.00	—	128K	$0.00
DeepSeek V4 Flashpinned Ultra-fast and cost-efficient reasoning and chat model.	$0.440	$1.32	—	1M	$0.00

Rates come from LiteLLM's price list. They are list prices: they exclude batch discounts, committed-use rates, priority tiers, and per-request surcharges such as web search. Cache-write cost is not modelled. A first pass over a cached prefix bills above the input rate, so a cold run costs more than shown here.

About this tool

01 · Pricing Catalogue

Aggregates over 2,400 models across 80+ providers from LiteLLM's unified index. Statically compiled to eliminate external network requests.

Units: Standardized USD / 1M input, output, and cached tokens.

02 · Segment-Weighted Estimator

Runs entirely in-tab with zero prompt leakage. Uses character-class weighting (code, CJK, digits, whitespace) to outperform naive chars/4 heuristics.

Precision: ±10% on prose, ±15% on code structures.

Operational Limits & Assumptions

· Budgeting vs Billing: Ideal for architectural sizing and price projection; not for final invoice reconciliation.
· List Pricing: Excludes volume tier discounts, enterprise commitments, and tool call add-on fees.
· Cache Reads: The cache slider applies published cache-read rates and excludes one-time cache-write costs.
· Snapshot Date: Reflected date indicates the price list sync timestamp.

Found a case it gets wrong? Tell me about it .