Cheapest AI APIs in 2026: Complete Price Ranking

Complete Ranking by Input Cost

Every major AI API ranked from cheapest to most expensive input price. Input cost matters most for tasks that process large documents, long conversations, or image analysis.

Complete Ranking by Output Cost

Output cost dominates when the model generates text — writing code, drafting emails, or producing summaries. Here are all models ranked by output price.

Cost Per Word

One million tokens is roughly 750,000 English words. The table below shows how much each model charges per 1,000 words of input and output — a more intuitive unit for writers, researchers, and content teams.

1,000 words ≈ 1,333 tokens. Costs shown are for standard-context pricing.

Cost Per Page

A typical page of text is around 500 words, or roughly 667 tokens. This table shows what it costs to process or generate a single page with each model.

Based on 667 tokens per page (500 words at ~1.33 tokens/word).

Cost Per Image Analysis

Analyzing a single image with a vision-capable model uses roughly1,000 input tokens. Here is what each model charges to process one image.

Based on ~1,000 tokens per image. Actual token count varies by image resolution and detail level.

Best Models for Budget-Conscious Developers

Not sure where to start? Here are three picks for different needs, each showing what $10 of output buys you.

How Much Can You Save?

The spread between the most and least expensive models is enormous. Switching from Claude Fable 5.1 to GPT-6 Luna saves you:

99.0% on input · 99.0% on output

Processing 1 million tokens of input costs $10 on Claude Fable 5.1 but only $0.10 on GPT-6 Luna.

For output, that same million tokens costs $50 vs. $0.50 — a 100x difference.

Of course, cheaper models aren't always a drop-in replacement. But for routine tasks like classification, summarization, and data extraction, budget models often perform just as well.