TPToolpazar

Global Araç

Ai Token Counter

Characters
322
Words
53
~ Tokens
85
Bytes (UTF-8)
325
ModelContextInput $/MOutput $/MEst. cost (1 call)
GPT-4o128,000$2.50$10.00$0.0052
GPT-4o mini128,000$0.15$0.60$0.0003
GPT-4 Turbo128,000$10.00$30.00$0.0158
Claude Opus 4200,000$15.00$75.00$0.0388
Claude Sonnet 4200,000$3.00$15.00$0.0078
Claude Haiku 4200,000$0.80$4.00$0.0021
Gemini 1.5 Pro2,000,000$1.25$5.00$0.0026
Gemini 1.5 Flash1,000,000$0.07$0.30$0.0002
Llama 3.1 70B128,000$0.35$0.40$0.0002
Mistral Large128,000$2.00$6.00$0.0032

Prices are list rates per million tokens. Estimate uses ~3.8 chars/token — within ~10% for English prose. For billing-critical counts, use each provider's official tokenizer.

Estimate tokens, characters, and API cost for GPT-4o, GPT-4, Claude, Gemini, Llama, and more — before you hit send.

Nasıl Kullanılır

  1. Paste the text.
  2. Set assumed output tokens.
  3. Read the per-model cost estimate.

Sık Sorulan Sorular

What's a token?

The atomic unit LLMs process. Roughly 3.8 characters per token in English, 1 word per ~1.3 tokens. Spaces, punctuation, and unusual words all count. Code and non-English text tokenize less efficiently (often 2-3x more tokens per character).

Why does my token count vary between models?

Different model families use different tokenizers. GPT-4 uses BPE (cl100k); Claude uses a different vocabulary; Gemini uses SentencePiece. Identical text produces 10-20% different token counts between them. Always estimate with the tokenizer matching your target model.

How accurate is this token estimate?

Within ~10% for English prose. Drifts 20-40% for code (tokens per character drops), non-English text (multi-byte characters are heavier), and numbers/symbols. For billing-critical workloads, use the model vendor's official tokenizer (tiktoken for OpenAI, Anthropic's library for Claude).

How many tokens fit in a context window?

GPT-4o and Claude Opus 4 handle 200k tokens (~150k words, ~600 pages of text). Gemini 1.5 Pro handles 2M tokens (~1.5M words, ~6000 pages). Most applications don't need more than 10-50k tokens per request. Context windows are about capability ceiling, not typical usage.