Pricing data
Pricing methodology
The public snapshot covers 1141 priced models across 131 providers.
What we track
Tokenscost records LLM API input and output prices in USD per million tokens, context windows and documented cached and batch rates. Provider-directory coverage is distinct from priced-model coverage. Missing fields mean not recorded, not free. Benchmark quality and token prices measure different things.
Sources and verification cadence
The live pricing table synchronizes stored model records daily; public file snapshots update when the site is rebuilt. Synchronization is not independent provider verification. The public snapshot was last updated 2026-10-31. Source links refer to pricing pages already recorded in our dataset; they are not evidence of a fresh independent audit. A model-specific last-verified date is shown only where recorded. Please report discrepancies through our contact page.
Blended and per-request costs
Request cost = (input tokens × input price + output tokens × output price) / 1,000,000. Our answer summaries use 1,000 input and 500 output tokens; calculators retain their selected workloads. Blended price is a token-weighted average, not the simple mean of rates. Multiply request cost by request volume for a budget; taxes, currency conversion, tools and hidden reasoning tokens may add charges.
Cached and batch handling
Cached input rates follow the recorded model/provider rules. Cache writes, storage, minimum prefix lengths and expiry can add costs. Batch rates use the recorded batch multiplier for both input and output. We do not assume that cache and batch discounts stack. Long-context tiers apply according to the recorded threshold. Not-recorded discounts are excluded from estimates.
GPU collection
The GPU snapshot lists USD per hour, instance, GPU count, region and tier with provider source links. Its recorded refresh cadence is weekly; the rental-history index uses monthly observations. Rates can exclude storage, networking, taxes and availability constraints. Multi-GPU instances should be normalized per GPU before comparing unit costs. Dataset timestamps are not refreshed just because a page is opened.
Changelog, license and attribution
Documented price changes are listed in the pricing-history and changelog pages. The pricing dataset is released under Creative Commons Attribution 4.0 (CC-BY-4.0). You may reuse it commercially or in research with attribution and a link to Tokenscost. Identify your accessed snapshot and distinguish observed prices from your own estimates. The license applies to the dataset, not third-party trademarks.
How to cite
Tokenscost (2026). LLM API pricing dataset. https://tokenscost.com/data. Accessed 2026-10-31.
Last updated: · Source: Pricing dataset · Methodology
Frequently asked questions
How are token costs calculated?
Multiply input tokens by the input rate and output tokens by the output rate, add the two amounts and divide by 1,000,000. A standard request example uses 1,000 input and 500 output tokens. Monthly budgets multiply the result by request volume; taxes, hidden reasoning tokens, tools and tier surcharges can add costs.
Which recorded model has the lowest request cost?
lfm-2.2-6b by Liquid costs $0.00002 for the standard request, with $0.01 input and $0.02 output per million tokens. This price ranking does not establish quality equivalence; verify task success and context requirements.
Can cached and batch discounts be combined?
Only recorded rates are used. Cached input can have cache-write or storage charges, while batch rates apply only to eligible asynchronous jobs. We do not automatically stack the discounts or treat an absent rate as zero. Check the linked provider source before budgeting.
How current is the public dataset?
The live pricing table synchronizes stored model records daily; public file snapshots update when the site is rebuilt. Synchronization is not independent provider verification. It contains 1141 priced models across 131 providers, last updated 2026-10-31. Directory entries are counted separately. A synchronization date is not proof of independent verification.