LLM API Cost Calculator

For raw price per token, the small models: GPT-6 Luna and Claude Haiku 5.5 at $0.10 per million input tokens, then Qwen3.8 Flash, GPT-5.6 Luna and DeepSeek V4.1 Flash. The ranking changes with the shape of the workload, though. Output-heavy work favours models with cheap output, long cached prompts favour models with cheap cache hits, and very long prompts can push some models onto long-context rates. Enter your own numbers and the chart ranks every model on that workload.

Use about 4 characters or three-quarters of a word per token for English prose, so a 1,500-word document is roughly 2,000 tokens. Code and non-English text usually take more. Count the whole request, not just the user's message: the system prompt, conversation history and tool definitions are resent on every call. For exact figures, every provider returns token counts in the API response, and some offer a token-counting endpoint.

No. Each model family has its own tokenizer, so identical text can produce noticeably different counts. Anthropic notes that Claude models from 4.7 onward use a tokenizer that produces about 30% more tokens for the same text than earlier ones. For a fair comparison, run a sample of real prompts through each provider and read the usage figures they return.

Whenever nobody is waiting on the answer: nightly summaries, back-filling tags, evaluations, bulk extraction. Anthropic, OpenAI and Google take 50% off input and output for batch requests, which return within 24 hours and often much sooner, and the discount stacks with caching. xAI takes 20% off Grok 4.3 and does not batch Grok 4.7. DeepSeek has no batch API but halves every rate off-peak, and Alibaba's Qwen models listed here have no batch discount.

On some models. OpenAI bills the whole request at twice the input and cache rates and 1.5× the output rate once the prompt passes 272,000 tokens. Gemini 3.1 Pro and both Grok models charge more from 200,000, Qwen3.7 Plus from 256,000, and Claude Haiku 5.5 from 100,000. Other Claude models, DeepSeek and Qwen3.8 bill their whole window at one rate. The calculator switches rates automatically and flags the models affected.

Server-side tools such as web search carry per-use fees, Anthropic's US-only inference costs 1.1× the standard rate, and fast or priority modes cost more. Negotiated and committed-use discounts lower it. This calculator covers the standard rates, which are the right baseline for comparing models.