LLM API pricing statistics, October 2026

Key figures

• On 12 October 2026, 338 AI models from 49 providers have a public API price per token.
• The median model costs $0.45 per million input tokens and $1.60 per million output tokens.
• The cheapest paid model is Llama 3.1 8B Instruct from Meta, served by DeepInfra, at $0.02 in and $0.04 out per million tokens.
• The most expensive is o1-pro from OpenAI, at $150.00 in and $600.00 out per million tokens.
• Output tokens cost 4.0 times as much as input tokens on the median model.
• 65 percent of paid models, 221 of 338, cost under $1 per million input tokens.
• 76 models have a cheaper batch price for jobs that can wait.
• Cached input is 90 percent cheaper than fresh input on the median model that offers it, and 214 models do.
• The largest context window is 2M tokens, on Grok 4.20 Multi-Agent. 107 models accept one million tokens or more.
• 31 models were listed in the last 30 days.

Models by input price

By provider

Providers with at least three paid models on the list.

How the figures are worked out

Each price is the model maker's own price for its standard service, as OpenRouter lists it. A promotional discount on it counts, since it is what you pay today, and the regular price is noted under the model. Faster or priority tiers cost more and flex tiers less, and both are left out. Where the maker does not serve a model itself, such as most Llama models, the price is the cheapest regular price among the hosts that do, and that host is named under the model. Short-lived discounts from other hosts are left out. Medians are taken across paid models only. Cheapest and most expensive are ranked on a blended price that weights input three to one over output, a typical mix for agent workloads. Every model and its price is in the full tracker.

How to cite

Quote any figure with a link to this page and the date, for example shurco.ai LLM API pricing statistics, 12 October 2026. The figures change every morning, so the date matters.

For AI agents and developers

The same figures are available as JSON, and every model's price as JSON. An agent can also call the MCP server at https://shurco.ai/mcp with no key, to look up a price, estimate a bill or find the cheapest model that fits.