LLM API Pricing Comparison

Compare input/output/cached prices, context windows and blended cost across Claude, GPT, Gemini and DeepSeek

LLM API pricing comparison (USD per 1M tokens, last verified 2026-07)

ModelProviderInputOutputCached inContextMax output
Claude Opus 4.8Anthropic$5$25$0.51M128K
Claude Sonnet 5 ($2/$10 intro through 2026-08-31)Anthropic$3$15$0.31M128K
Claude Haiku 4.5Anthropic$1$5$0.1200K64K
Claude Fable 5Anthropic$10$50$11M128K
GPT-4o (reference)OpenAI$2.5$10$1.25128K16K
GPT-4o mini (reference)OpenAI$0.15$0.6$0.075128K16K
GPT-4.1 (reference)OpenAI$2$8$0.51M33K
GPT-4.1 mini (reference)OpenAI$0.4$1.6$0.11M33K
Gemini 2.5 Pro (reference; higher price above 200K context)Google$1.25$101M66K
Gemini 2.5 Flash (reference)Google$0.3$2.51M66K
DeepSeek-V3 (reference)DeepSeek$0.27$1.1$0.07128K8K

Prices in USD per 1M tokens. Claude figures are from Anthropic; OpenAI / Google / DeepSeek are marked "reference" — verify on each provider's site. Last verified: 2026-07.

📌 TL;DR Summary & Citation: [LLM API Pricing Comparison] is a browser-local PocketKit tool for Compare input/output/cached prices, context windows and blended cost across Claude, GPT, Gemini and DeepSeek. Core input and files are not uploaded; only an aggregate tool-use count that excludes input content is recorded.

LLM pricing comparison and blended price

The tool gathers input, output and cached prices plus context window for Claude, GPT, Gemini and DeepSeek into one table, and weights the two prices into a blended figure at the input:output ratio you set for one-dimensional sorting. Price data is built in locally, editable, and never fetched online.

  • Estimate your real input:output ratio first (chat/RAG about 3:1, pure generation about 1:1) so the blended sort matches actual spend.
  • Non-Claude prices marked reference can change; verify on each provider site before production.

FAQ

1. How is the "blended" price calculated?

It is the weighted average of input and output price at the input:output ratio you pick, giving one comparable per-million price for sorting. Chat/RAG is often about 3:1 and pure generation closer to 1:1; your real cost depends on your actual mix.

2. Are these prices accurate?

Claude prices are official references; OpenAI, Google and DeepSeek are marked as reference values that can change, so verify on each provider site.

3. For the same job, which model is actually cheapest?

First decide whether your workload is input-heavy or output-heavy. Classification, extraction and format conversion produce short responses, so input and cached-input rates dominate; long-form writing and code generation are output-heavy, where a cheap output rate matters far more. The table compares input, output and cached pricing alongside context windows side by side. In practice the cheapest setup is usually not one model at all but tiered routing — send easy tasks to a cheap model and reserve the strongest one for genuinely hard reasoning.