A newsletter de dados da OpenRouter

Inferred

Novas análises sobre o uso de tokens de IA, extraídas da atividade agregada em todo o OpenRouter.

Todas as edições

Share of family requests on an older version, and share of accounts using one.

Inferred 05Mais recente23 de out. de 2026

Version lag

This week - version lag. Or put another way, why the hell are some people still using Claude Sonnet 4.5?

A quarter of requests run on an older version
Nearly a third of accounts use an outdated model
Latest isn’t always cheapest

Também nesta edição

Latest version share of family requests, by days since launch.

Change in cost if older-version traffic moved to the latest version, at list prices.

Models used per month, by percentile, AI-native startups and large orgs.

Inferred 0415 de out. de 2026

Model portfolios

This week - a deep dive into AI-native startups and large organizations on OpenRouter. How many models do they use, how much of the work goes to their main model, how often do they swap that main model out, and which tasks hold on to a model the longest.

Portfolios grew by 40% in a year
One model still does two-thirds of the work
Switching the main model is routine
Specific workloads switch at different rates

Também nesta edição

Average models used per month against the request share of the top model.

Share of orgs switching their top model, and switching to one from a different lab.

Share of accounts switching their top model for a given task, on average each month.

Share of weekly tokens by token type and user type.

Inferred 038 de out. de 2026

Cache is Crucial

As agents take over the token stream on OpenRouter, cache rate is fast becoming the bellwether for performance.

Agents are the primary token users
Agentic tokens are mostly cached
Cache optimization has happened industry-wide

Também nesta edição

Weekly tokens by user type on OpenRouter.

Average cache rate for agentic workloads by quarterly model cohort.

Agentic cache rate by inference provider, January 2026. Bubble size is prompt tokens.

Agentic cache rate by inference provider, September 2026. Bubble size is prompt tokens.

Number of public models launched on OpenRouter by month.

Inferred 021 de out. de 2026

Model Half-Life

How long should we expect a model to hold users? Launches are accelerating and retention is falling.

Launches have accelerated
Retention is falling

Também nesta edição

Percent of requests by task that went to a model released in the last 60 days, as of September 30, 2026.

Weekly share of spend on OpenRouter between Anthropic and OpenAI models, last six months.

Retention at weeks 9 to 12 for users that tried a model from each lab in a given year.

Weekly share of spend between OpenAI and Anthropic models on OpenRouter.

Inferred 0122 de set. de 2026

The Token State of Play

OpenAI, Anthropic, and the rise of everyone else.

OpenAI recently caught up to Anthropic in platform spend
Token growth shows no signs of fatigue

Também nesta edição

Weekly tokens used on OpenRouter, January 1, 2025 to September 20, 2026.

Weekly share of tokens between OpenAI and Anthropic models on OpenRouter.

Weekly share of spend by model between Anthropic and OpenAI, 2026 so far.

Weekly share of tokens by model between Anthropic and OpenAI, 2026 so far.