LLM Pricing Leaderboard

Shaded rows are prices we could not confirm on an official provider page in the latest research pass (dataset verified 2026-10-20). Tiered, long-context, batch and peak variants live in each model's notes — the cost calculator surfaces the ones that change a bill, including DeepSeek's 2× peak rates and dated promotional increases.

Reading the table

Three cautions before you rank models by a single column. First, headline rates have conditions: several models price by prompt length, long context costs more, and DeepSeek's rates double in peak hours — the calculator's footnotes carry the details per model. Second, cached input is real money: if your workload repeats a long system prompt or document prefix, the cached-input column can matter more than the headline input rate, cutting that portion by 75–99% on some models. Third, promotional prices expire: introductory rates with announced end dates are on this table at today's price because that is what you would pay today — the calculator flags the scheduled rise so a decision made this month still makes sense next quarter.

To turn any row into an actual budget, take the model to the LLM API cost calculator and run your own request volume and token averages through it — the cheapest model per token is frequently not the cheapest model per month once output length and caching enter the picture.