LLM API pricing comparison

· Fixed usage: 100 calls/day · 2,000 input / 800 output tokens per call · 30-day month;
· Off-peak basis: time-based models are priced at the off-peak rate (DeepSeek off-peak ≈ half of peak; peak window weekdays 9:00–12:00, 14:00–18:00);
· Cache: 30% hit rate, effective input price = 30% × cache-hit price + 70% × miss price; models without a published cache price use the miss price;
· Tiers: input-length-tiered models are priced at the tier a 2,000-token input falls into;
· Currencies: unit prices keep the official currency (CNY or USD per 1M tokens); monthly costs are converted to CNY at the reference rate for ranking — indicative only;
· Step-by-step derivation in How to calculate LLM API costs; conventions in the pricing methodology.

· Output price dominates the bill: at the example usage, output contributes about 55–65% of the monthly cost (800 output tokens × output price); comparing input prices alone systematically underestimates;
· Cache prices differ by orders of magnitude: the two DeepSeek models' cache-hit prices (¥0.02/1M off-peak) are about 50× lower than their miss input price (¥1/1M off-peak) — at high hit rates, input cost approaches the cache price;
· Off-peak half price is a predictable gain: time-based models are estimated at the off-peak rate; moving batch jobs off-peak halves the cost by billing rule, not by assumption;
· Long-context tiers are never back-filled: GPT-6 series >272K rates are unpublished, so estimates in that range show "cannot estimate" instead of borrowing the lower tier's price.

· Data comes from this site's single source models.json; every record carries an official source link; verification dates appear in the table and on each model profile;
· The footer aggregates all official sources; price-verification time ≠ page-content update time: re-verification without price changes does not update page or sitemap dates (methodology §6);
· Cross-currency conversion uses the CFETS central parity rate (a static reference rate, not a trading rate);
· Estimates exclude free tiers, resource packs, subscription discounts and limited-time promotions; they are not provider invoices and do not constitute purchase advice.

· Recompute with your real usage: cost calculator (any combination of models side by side);
· Full pricing profile per model: model library;
· Two candidates line by line: model comparison, e.g. GLM-5.3-Flash vs GPT-6 Luna (cross-vendor lightweight);
· Scenario-based monthly costs: cost scenarios guide;
· Back to the guide directory.