LLM API Prices tracks 50 popular models, including 47 verified first-party API prices and 50 OpenRouter matches. Use this hub to move from the full interactive calculator into focused, source-linked pricing guides without losing the distinction between direct and routed API costs.
What can you compare in these pricing guides?
The pricing hub is organized by user intent rather than by the internal shape of the dataset. Use the cost ranking when price is your first constraint, a provider guide when you already know the API family, and a cross-provider comparison when you are still deciding which ecosystem to adopt.
Cheapest LLM APIs
Rank verified APIs by output-token price, compare OpenRouter rates, and understand why input/output ratios can change the winner.
Anthropic API Pricing
Compare Claude fresh input, cached input, output, and an identical example workload across the current model ladder.
OpenAI vs Anthropic vs Gemini
Apply one workload to representative budget, balanced, and premium tiers from three major model families.
LLM API Price History
See the verified baseline and every future input, cached-input, or output price change recorded for official and OpenRouter channels.
Browse provider and model pricing pages
Use a provider directory when you already know the ecosystem, or open a model page when you need one answer about input, cached-input, output, context, source, and example workload cost.
Popular starting points: GPT-5.6 Sol · GPT-5.6 Terra · GPT-5.6 Luna · GPT-5.4 · GPT-5.4 mini · GPT-5.3 Codex.
How should you read an LLM price table?
Input price covers the prompt and context processed by the model. Output price covers generated tokens and is often the larger rate. Cached-input price applies when a provider recognizes reusable prompt content under its caching rules. Context-window size describes capacity, not a flat fee: you pay for the tokens actually processed, subject to any long-context tier.
A useful comparison therefore starts with a workload, not a model name. Estimate monthly fresh input, cached input, generated output, and request count. Then add channel-specific fees and any known regional, tool, storage, or priority-processing charges. The homepage calculator applies that structure consistently across every verified model.
Why do official and OpenRouter prices appear separately?
Official direct pricing comes from the first-party model provider. OpenRouter is a separate routing and billing channel that can expose the same model through one API, sometimes with multiple underlying endpoints. The base inference rate may match the official price, differ from it, or exist even when the model maker does not sell a first-party hosted API.
LLM API Prices never replaces an official price with an OpenRouter price. Each channel keeps its own model ID, source, verification time, and token rates. OpenRouter’s credit purchase fee is also shown as a separate optional calculator item instead of being hidden inside the inference price.
How are these static pricing pages kept current?
The catalog is stored as structured JSON, but the important guide text, price tables, source links, headings, and answers are generated into the raw HTML. Search engines and AI answer systems can read the core content without executing JavaScript. After the daily source check, the site rebuilds the Hub, comparison pages, homepage snapshot, and Sitemap from the latest validated data.
Automatic publication is deliberately limited. A missing source keeps the last-known-good record. A suspicious official price change is held for review. OpenRouter model identity uses a curated mapping rather than fuzzy name matching, which reduces the risk of comparing similar-sounding but different model versions.
LLM API pricing guide questions
What does LLM API Prices include?
LLM API Prices stores input, output, cached-input, context, source, and verification data separately. Official direct prices and OpenRouter prices are shown as different channels rather than combined into one number.
How often are the pricing guides updated?
The scheduled job checks pricing sources daily at 09:00 Asia/Shanghai. Safe machine-readable updates can rebuild the static pages automatically, while suspicious or structurally changed official prices are held for review.
Are the models in each comparison equivalent?
No. Price tables compare published cost dimensions, not capability equivalence. Model quality, latency, tools, context behavior, rate limits, and data policies must be tested for the intended workload.
Why separate official pricing from OpenRouter?
The two routes can have different model identifiers, routing behavior, platform fees, account terms, and availability. Keeping them separate lets users understand exactly which channel a price belongs to.