Pricing Comparison of Top LLM APIs (Cost per 1M Tokens)

Token costs are the primary variable expense for AI products. For any application built on hosted LLM APIs, the per-token rate multiplied by usage volume defines gross margin — making pricing tables required reading for founders, CFOs, and engineering leads alike.

The market now offers substitutes at every tier. With five providers — OpenAI, Anthropic, Google, xAI, and DeepSeek — fielding 13 models across flagship, mid, mini, and nano classes, buyers can arbitrage capability against cost in ways that were impossible when only one or two frontier models existed.

A single comparable metric cuts through the noise. Providers quote input and output prices separately, which obscures true cost. Combining 1M input + 1M output tokens into one number gives a level playing field for ranking — even if individual workloads skew heavily toward one side.

OpenAI's GPT-5.5 sits at the top of the ranking at $35 combined per 1M input + 1M output tokens, driven by a $30-per-million output rate — six times its $5 input rate.

Anthropic's Claude Opus 4.8 matches GPT-5.5 exactly on input pricing at $5 per 1M tokens but undercuts it on output at $25, landing at $30 combined. That $5 gap is small in percentage terms but meaningful at scale: across a billion output tokens, it amounts to $5,000 in savings.

These flagship rates position top-end reasoning as a premium product — one that makes economic sense for high-value tasks like complex analysis, agentic workflows, and code generation, but not for routine, high-volume text processing.

At the other end of the chart, GPT-5.4-nano charges just $0.20 per 1M input tokens and $1.25 per 1M output tokens — $1.45 combined, or roughly 1/24th the cost of GPT-5.5.

One step up, GPT-5.4-mini lands at $5.25 combined ($0.75 input / $4.50 output), still less than a sixth of flagship pricing while occupying the same product family as OpenAI's mid-tier GPT-5.4 ($17.50 combined).

This tiering creates a clear laddering strategy: providers give developers a flagship for hard problems, a workhorse for standard tasks, and near-commodity tiers for classification, extraction, and routing — the kinds of jobs that generate the largest token volumes.

LLM API pricing now spans roughly 24x from top to bottom — $35 for GPT-5.5 down to $1.45 for GPT-5.4-nano per combined 1M input + 1M output tokens — making model selection the single largest cost lever in AI development.

The flagship race is tight: Claude Opus 4.8 ($30 combined) trails GPT-5.5 ($35) by $5, with identical $5 input pricing, so the real differentiation at the top happens on output rates and capability, not input cost.

Output tokens carry a consistent 5-6x premium over input tokens, so real-world bills depend heavily on how much a model generates — not just how much it reads.

For buyers, the winning strategy is routing, not loyalty: match each task to the cheapest tier that meets its quality bar, and revisit the pricing table often — with five providers competing across 13 models, these numbers move fast.