Watch OpenRouter for Price Rises and Retirements on the Models You Use

An agency builds chat features for a dozen clients and routes them all through OpenRouter. One client runs on Claude Sonnet, two on GPT-4o-mini because they’re cheap, a few on Gemini Flash for the long context. The agency quotes fixed monthly prices. A price change on any of those models comes straight out of its margin, and a retirement date means a client project to schedule.

OpenRouter publishes its whole model list with prices, limits and retirement dates, and it needs no key. That makes it the easiest list to watch. RadarModel AI reads it every hour from a small server the agency already has, and posts to the team’s Slack channel when one of the watched models changes.

Four steps, from listing the models routed to through OpenRouter to acting before a change bites, next to a report with a context cut as breaking, a price rise and a retirement date as warnings, and a new model as info.

Prices, limits and retirement dates, from the list itself.

The setup

{
  "interval": "1h",
  "providers": [{"type": "openrouter"}],
  "watch": [
    "openrouter:anthropic/claude-sonnet-4.5",
    "openrouter:openai/gpt-4o-mini",
    "openrouter:google/gemini-2.5-flash"
  ],
  "alerts": {"min_severity": "warning", "slack": {"url_env": "SLACK_WEBHOOK_URL"}}
}

radarmodel-ai watch keeps it running. A * in the watch list works too: openrouter:anthropic/* covers every Anthropic model on OpenRouter.

A week where three things move

The run below is the released binary on a saved copy of the list, edited so that three of the watched models change at once and a new model appears:

$ radarmodel-ai check -min info
RadarModel AI: 1 breaking, 2 warnings, 1 info
openrouter  BREAKING  anthropic/claude-sonnet-4.5 *  context_length 1000000 -> 200000
openrouter  WARNING   google/gemini-2.5-flash *      input_price 0.0000003 -> 0.00000035
openrouter  INFO      mistralai/mistral-medium-3.5   new
openrouter  WARNING   openai/gpt-4o-mini *           expires now 2026-12-31 (scheduled to be retired)

Each line is something the agency can act on. The shorter context window breaks the client whose documents run past 200,000 tokens, so that’s breaking. The price rise is a warning: the margin on those clients just got thinner, and it’s time to check the quote. The retirement date gives the GPT-4o-mini clients until the end of the year to move. The new Mistral model isn’t watched, so it’s info, a line for whoever likes to try new models.

Slack gets the same three watched changes, once. The next hourly scan is quiet unless something else moves. The reference has the full table of what counts as what.