LLM API cost calculator.

Providers bill per token, with separate prices for input (your prompt, system prompt, retrieved context and chat history) and output (the model's reply). Monthly cost = requests × average input tokens × input price + requests × average output tokens × output price, with prices quoted per 1 million tokens. Reasoning or “thinking” tokens are billed as output, so count them in your output average.

For English text, one token is roughly ¾ of a word, so 1,000 tokens ≈ 750 words and a typical page of text is 500–700 tokens. Code, non-English text and JSON usually take more tokens per word. Each provider uses its own tokenizer, so the same text can count differently across models; use the provider's token counter for exact numbers.

Cached input tokens are billed at a fraction of the normal input price: about 90–98% off on most of the models listed here. Caching only applies to a repeated prompt prefix (a long system prompt, tool definitions, a shared document), so the cache-hit slider should reflect the share of input that is identical across requests. Anthropic also charges a one-off premium (1.25× input) to write a 5-minute cache entry; this calculator models cache reads only, which dominates once the cache is warm.

Anthropic, OpenAI and Google all offer an asynchronous batch API that processes requests within 24 hours at 50% off both input and output tokens. It suits work nobody is waiting on: nightly enrichment, evaluations, back-filling embeddings or summaries, bulk classification. Batch and caching discounts stack. The open-weight prices shown here (Together AI) have no published batch rate, so batching doesn't change them.

A few models charge higher rates once a single prompt passes a size threshold: Claude Haiku 5.5 above 100K input tokens, Gemini 3.1 Pro above 200K, and GPT-6 Astra above 272K. The calculator switches to those rates automatically when your average input is over the threshold and flags the row.

Prices were last checked against each provider's official pricing page on November 2, 2026, at standard pay-as-you-go rates in USD. Two Gemini models are on introductory prices: Gemini 4 Argon until November 30, 2026, and Gemini 3.8 Flash until December 31, 2026. Their rows show the price that follows. Enterprise discounts, provisioned throughput, regional or data-residency surcharges and cloud-marketplace pricing (Bedrock, Vertex AI, Azure) can differ.

No. Tokens are usually the largest variable cost, but production AI features also pay for embeddings and a vector database, retries and evaluation runs, observability, guardrails and the engineering time to keep prompts, models and costs tuned. Use the AI build cost calculator for the one-time cost of building the feature.