Multi-Model Token Counter
Count tokens and estimate cost across GPT, Claude, Gemini, Llama and DeepSeek side-by-side.
| Model | Input tokens | Context | Input cost | Output cost | Total |
|---|---|---|---|---|---|
GPT-5Exact openai | 0 | 0.00% | $0 | $0.0075 | $0.0075 |
GPT-5 miniExact openai | 0 | 0.00% | $0 | $0.0006 | $0.0006 |
GPT-4oExact openai | 0 | 0.00% | $0 | $0.0050 | $0.0050 |
Claude Opus 4.6Approx anthropic | 0 | 0.00% | $0 | $0.037 | $0.037 |
Claude Sonnet 4.6Approx anthropic | 0 | 0.00% | $0 | $0.0075 | $0.0075 |
Claude Haiku 4.5Approx anthropic | 0 | 0.00% | $0 | $0.0020 | $0.0020 |
Gemini 3 ProApprox google | 0 | 0.00% | $0 | $0.0050 | $0.0050 |
Llama 4 405B (host)Exact meta | 0 | 0.00% | $0 | $0.0014 | $0.0014 |
DeepSeek V3Exact deepseek | 0 | 0.00% | $0 | $0.0006 | $0.0006 |
About these counts
Exact rows use the model family's actual tokenizer running in your browser:o200k_base (gpt-tokenizer) for GPT-4o/5, and Llama-3's tokenizer for Llama 4 (DeepSeek uses Llama-style BPE, marked approx).
Approx rows: Anthropic hasn't released a tokenizer for Claude 3+, so Claude rows use the Claude-2 vocab — expect ~5–10% drift. Gemini's tokenizer isn't public at all, so it uses a character-based estimate that's typically within 10–15% for English prose.
Pricing snapshot 2026-05-16. Costs are USD per token at list pricing; volume discounts and cached input pricing aren't reflected.
All tokenization happens in your browser. Your prompt never leaves your machine.
About this tool
Different LLM providers use different tokenizers. The same prompt can cost very differently across models — and counts you see in one provider's playground rarely match another's. This tool runs the actual tokenizers in your browser and shows accurate token counts plus per-1K-token costs for every major model.
All counting happens locally. Your prompt never leaves your browser.