Pıer
TidesCurrentsHarbor LightsLabBottlesAshore
Pıer

Navigation

  • Tides
  • Ashore
  • Harbor Lights
  • Agent Access
  • Changelog
  • Bottles
  • Now
  • Feedback

External links

GitHubCloudborne ↗

© 2026 Pier.

Methodology

Inclusion & scoring rules

Inclusion

Verified released foundation models appear on one default board, including those without evaluations or pricing. Vendor-official previews have a separate view; rumors and unannounced models are excluded. Each foundation model occupies one row, with configurations and snapshots in its details. Unknown names require identity verification, not fuzzy automatic merging.

Ability (consensus percentile)

Each source's scores are first converted to tie-aware 0–100 percentiles within that source. When a source lists multiple configurations, the source-designated default is used if available, otherwise the median — never the best configuration. Multiple boards run by the same operator are averaged within the group first, then independent operators are combined with equal weight. A missing source never counts as zero (it does not directly lower ability); it only reduces confidence.

Confidence

Confidence combines coverage (45%), freshness (25%), agreement (25%), and identity mapping (5%), independently of ability. Limited coverage does not exclude a model. Without usable evaluations, ability, confidence, and rank display —.

Value

Every included model remains visible. Comparable prices use official USD global or international standard online, uncached, ordinary-context rates; time-of-day pricing uses the peak rate. The aggregate usage basket is 1M input + 0.25M output tokens. Value is the geometric mean of ability percentile and inverted-cost percentile. Non-USD rates retain their original currency without FX conversion. Unknown prices, subscriptions, self-hosting, and non-comparable plans are not scored and display —.

Refresh & failure handling

Failed assessment sources retain their last valid snapshot. If all assessments fail, no new ability scores are published. Unchanged scores do not advance the score-change clock; history retains one effective point per day. Prices are checked independently. Failures preserve the last valid quote and its actual verification date, never extending a promotion. Price changes update value only, not ability scores or ability history. Newly listed models show no rank movement.