Model pricing guide
Compare AI Model Pricing for Token Costs
The cheapest model is not always the cheapest workflow. Compare GPT, Claude, and Gemini prices with your actual input tokens, output length, retries, cache eligibility, and quality requirements.
Use the same workload for every model
Start with one representative user action: the prompt, system instructions, retrieved context, tool results, and expected answer. Estimate input tokens and output tokens once, then apply each model's current input and output price per million tokens.
Do not compare a short test prompt on one model with a full production prompt on another. Long system prompts, chat history, RAG chunks, and tool schemas can change the ranking more than the headline model price.
Separate input, output, and cached tokens
Most provider pricing pages separate input and output tokens, and output tokens often cost more. Some providers also price cached input, cache writes, batch jobs, or context storage differently. Keep those buckets separate so a repeated support bot, code reviewer, or document summarizer is not estimated from one blended token price.
Comparison example
A workflow sends 20,000 input tokens and generates 2,000 output tokens. A model with a low input price can still lose if its output price is high and the app writes long reports. A stronger model can also be cheaper overall if it needs fewer retry or repair calls.
Compare quality-adjusted cost
For routing, classification, extraction, rewriting, and draft generation, test lower-cost models first. For final review, complex reasoning, sensitive decisions, or high-value customer answers, compare the cost of a stronger model against the cost of failures, manual review, and repeated prompts.
A practical model comparison table should include price per 1M input tokens, price per 1M output tokens, context window, observed success rate, average retries, latency, and whether the model supports the tool use, structured output, or language quality your product needs.
Refresh prices before launch
Provider prices and model names change. Before a purchase decision, confirm the current numbers in the official OpenAI pricing page, Claude pricing page, and Gemini pricing page. Then update the editable price fields in the calculator and save the assumptions with your estimate.
Next step
If the same prompt prefix appears in many requests, compare the model prices with the prompt caching cost guide. If traffic arrives in bursts, check token rate limits and API capacity before choosing a provider only by price.