Equivalent API $ and Value ×
Default agent turn: 12,000 input tokens with 70% cache hit plus 1,500 output tokens. Unit prices prefer OpenRouter fair listings (snapshot 2026-08-19); models not listed fall back to the vendor's official API price (permanently discounted tiers at the rate users actually pay); when neither exists, the closest Artificial Analysis-scored model proxies the price and is flagged estimated. Credit-based plans (GLM / MiMo) convert credits to tokens with the vendors' official multiplier tables; peak/off-peak tiers quote the full-peak figure with off-peak as the upper bound. Quotas marked ~ are the vendor's rough figures. The vs Claude column assumes Claude Pro ≈ 500 turns/month (community estimate).
Equiv. API $ = monthly allowance in turns × official API price of the flagship model
Value × = Equiv. API $ / monthly list price
If a vendor only publishes a ratio such as 5× Pro, Value is marked ratio only. We do not invent a token count.
Quotas and parallel sessions
Quota fields stay split: token / credits / prompts / requests, and 5-hour / weekly / monthly windows. Parallel sessions are official when published, otherwise —. Community measurements are a separate column. Capacity / $ prefers official monthly tokens; if a vendor only publishes credits or a ratio, the chart uses tokens implied from Equiv. API $.
History
Each plan is a time series: list price, yearly effective, quota fields, parallel, models, and status. A sparkline on the leaderboard is the last 90 days. The plan page has the full axis. Every production snapshot must keep captured_at, source_url, and official | estimated | unknown.
This build
The public site is static HTML plus one shared, cacheable plans.json. data/catalog.py is the source. scripts/build.py renders one folder per language into dist/. Filters, charts, column toggles, and search run in the browser after a single catalog fetch.