Compare LLM API prices and find the best-value AI model

By per-million-token pricing, the floor in this catalog is Microsoft's Phi-4-mini at $0.02/$0.04 — though it is a small model with only 128K of context. For the cheapest option that still gives you a million-token window, Alibaba's qwen3.7-flash is $0.03/$0.13, with Inception's mercury-2.5-preview close behind at $0.04/$0.15 over 262K. Raise the quality bar to 90 and the cheapest is qwen3.5 at $0.10/$0.15. Output rates down here run more than a hundred times below the flagship tier, so sort the table above by price and cross-check the value score. Note that launch promotions expire: every row on this site carries its price source and verification date, so budget long-term against the list rate.

GPT-5.5 scores higher on multimodal and overall reasoning, ideal for flagship tasks mixing text, image and audio. Claude Sonnet 5 excels at coding and long-task reliability at a friendlier price. Favor Claude for code-heavy work and GPT-5.5 for an all-round flagship, then compare quality and cost on your actual tasks.

LLM APIs bill separately per million tokens for input (what you send) and output (what the model generates), with output usually several times more expensive. Monthly cost ≈ (avg input tokens × rate + avg output tokens × rate) × monthly requests. Use our cost calculator to estimate it directly.

RAG depends on both context window size and input token price, since retrieved documents consume a lot of input. Gemini's huge context plus low input pricing suits RAG well; for smaller corpora, a cheaper mid-tier model with good retrieval and caching keeps costs down.

Define your quality bar first, then compare input/output pricing, latency and speed, context length and modalities. Start from a scenario (coding, RAG, production), narrow down with the comparison table and recommendation wizard, and validate your monthly budget with the cost calculator.