TechCompare LogoTechCompare

How much do tokens cost? LLM pricing per million tokens compared across providers

Token costs are small per-request but add up fast at scale. Use prompt caching for repeated content, batch mode for non-urgent processing, and cheaper models for tasks that don't need frontier intelligence. The built-in LLM Pricing Calculator on this site estimates costs across models and usage patterns.

LLM API pricing is quoted per million tokens, split between input tokens (what you send) and output tokens (what the model generates). Output tokens typically cost 3-5x more than input tokens because generation is computationally more expensive. The cheapest models (DeepSeek V4 Flash, GPT-5.4 Nano) cost under $0.50 per million input tokens. The most expensive (Claude Opus 4.8) costs $5 per million input tokens.

By TechCompare · Updated

Knowledge area
Fundamentals
How tokens and tokenization work
Topic focus
Token cost guide
token-cost

How this is calculated

To calculate cost: (input_tokens / 1,000,000) × input_price + (output_tokens / 1,000,000) × output_price. A single ChatGPT interaction with a 500-token prompt and 200-token response on GPT-5.4 costs roughly $0.003. A high-volume application processing 1 million requests per day with 1,000 input tokens and 500 output tokens each on GPT-5.4 costs about $7,500 per day. Prompt caching dramatically reduces costs for repeated content: cached input tokens on GPT-5.4 cost $0.25 per million (90% discount). Batch mode (24-hour turnaround) gives a 50% discount. Always check if your use case qualifies for caching or batch pricing before scaling.

Verdict

Pricing splits between input and output tokens, with output usually 3-5x pricier because generation is compute heavy. The math is (input / 1,000,000) x input price plus (output / 1,000,000) x output price. A 500-token in and 200-token out GPT-5.4 chat costs about $0.003, but a million calls a day at 1K in and 500 out runs around $7,500 daily. Cached input on GPT-5.4 drops to $0.25 per million, and batch mode shaves another 50% for 24-hour turnaround.

More Tokens scenarios

What are tokens
A token is the atomic unit of text that a language model processes.
View details ➜
OpenAI vs Llama tokenizer
A sentence that costs 50 tokens on GPT-5 might cost 55 tokens on Llama 4 or 48 on Claude.
View details ➜
Token limits by model
Context window size is the maximum number of tokens a model can process in a single request, including both input and output.
View details ➜

Frequently asked questions

Why do output tokens cost more than input tokens?
Generation is the expensive part. Processing input tokens is a parallel operation over the whole prompt, while output tokens must be produced one at a time, re-running the model per token. That's why most providers price output at 3-5x input, and why trimming the answer length often saves more than trimming the prompt.
How do I calculate the cost of an LLM request?
Input tokens divided by a million, times the input price, plus output tokens divided by a million, times the output price. A 500-token prompt with a 200-token reply on GPT-5.4 runs about $0.003. Scale that to a million daily requests at 1,000 tokens in and 500 out and you're near $7,500 a day, which is where caching and batch discounts start to matter.
How much do prompt caching and batch mode save?
Caching typically cuts cached input tokens by 50-90% (GPT-5.4 cached input runs $0.25 per million versus $2.50 full price). Batch mode, with up to 24-hour turnaround, halves both sides. Combined, a workload that reuses a long system prompt and tolerates delayed results can run at a quarter of the naive bill.