About AI Wave
AI Wave is a trend dashboard that shows how AI models and tools are moving, in one view. It collects metrics from public sources every hour, computes a relative trend per model, and surfaces new releases or pricing and API changes with a link to the source.
The reason it exists is simple: models change so fast that a technology choice made yesterday may not be the best one today, and there was no single place to see that movement. Instead of checking each vendor's blog, release notes, repository and price list separately, this shows what moved and by how much first.
What is collected
Everything comes from public APIs that need no authentication. No scraping, no private data.
- OpenRouter model catalogue: input and output token rates. A change in rate creates a change event.
- GitHub: repository star history.
- Hugging Face: downloads and likes for open-weight models, plus new models published under a vendor's organisation.
- npm: weekly downloads of the provider's official SDK.
How momentum is calculated
The score is not an absolute measure of performance or quality. It is a relative indicator for comparing how much several public signals have moved recently.
- Each signal is compared against that model's own past observations. Observations are bucketed by day, so a high score goes to the model that moved more than usual, not to the model with the biggest absolute numbers.
- Signals are weighted and summed. A model missing a signal is not scored zero for it; that weight is redistributed proportionally across the signals that are available.
- Finally, the result is trusted only as far as data completeness allows. So that a model with one signal cannot outrank a model with all four, a thin evidence base pulls the score toward the neutral value of 50.
What this cannot tell you
Hiding this would lead you to misread the numbers, so it is stated plainly.
- Actual usage is unknown. Only providers know how often each model is called, and they do not publish it. With no free source available, this signal is excluded from the weights. Developer adoption (SDK downloads) is a separate signal, not a substitute.
- There is no search-interest signal. No official free API exists.
- Commercial models have fewer signals. Because their weights are not published, no Hugging Face signal is available. Their data completeness is therefore lower and their scores are pulled further toward the neutral value. Completeness is shown on each model page.
- Automatic detection produces false positives. Only entries confirmed by an official announcement carry the official badge; the rest are marked pending review. Please check the original source as well.
- These metrics and prices are not suitable as a basis for investment, purchasing or contractual decisions.
Update cadence
Metrics are collected hourly. Scores, however, are derived from day-over-day change, so a value appears only after at least three days of observations; before that the score shows as computing. Every view shows when the data was last refreshed.
Member features
The public dashboard and every detail page are available without an account. Only following models and receiving change alerts require signing in. What is collected and for how long is set out in the privacy policy. Privacy policy
Advertising
Google AdSense ads cover running costs. Ad areas are labelled as advertisements, and advertisers have no influence on how scores are calculated. See the cookie notice for details on cookies. Cookie notice
Languages
Korean and English are available. The initial language follows your browser settings, and you can switch at any time with the selector at the bottom of the page.
Contact
For questions about the service or data corrections, please reach out at the address below. elopadmin@elopstudio.com
If you find an incorrect metric or a false positive, please tell us. We will verify it, correct it and note the correction on that entry.