Checked 2026-10-15

How we compare plans

Every cell is filled. A number is either sourced and dated, or it's an estimateest. with its method one hover away.

1. Price tiers

Plans are grouped by monthly list price in USD for one individual seat on monthly billing: Free, ~$8–10, ~$20, ~$30–40, ~$50–80, ~$100 and ~$130–200. Team-only plans (Augment Business, Windsurf Teams) are marked as such. Gemini Code Assist Enterprise ($54/mo, no longer sold) and the Alibaba Token Plan appear only on /plans or not at all. Non-USD plans (Kimi's China pricing) use the vendor's international USD price.

2. Verified vs estimated

Verified means the number appears on the vendor's pricing page, docs or help center, or on a public benchmark leaderboard, checked on the date shown. Estimated (dotted underline, est., orange on /value) covers everything else. Each estimate carries a tooltip that names its sources and derivation. We use these kinds of estimates, in order of preference:

  1. Vendor statements outside the pricing page: staff posts on X (OpenAI's Codex lead: Pro 100 = 5x Plus, Pro 200 = 10x), archived help pages (Anthropic's 2025 prompt counts via archive.ph), earlier vendor tables (Windsurf's March 2026 messages-per-day table).
  2. Third-party captures of vendor pages: MiniMax call counts (iadecider.com, Sept 23), Z.ai Pro/Max prices (AI Compare Ninja), Amp's included usage (pricing trackers), Factory's earlier 20M-token allowance.
  3. Community measurements: Antigravity Pro (~8 Opus prompts per 5h, ~24 a week, Reddit) and Cursor (one user's ~$200 of API-priced tokens from a legacy $20 plan). These are low-confidence, and we say so.
  4. Closest-model proxies for benchmarks: when a plan's model isn't on a leaderboard, we show the score of the nearest benchmarked model and state which direction it likely errs (e.g. Kimi K3 for K2.7 Code is an upper bound; Opus 4.8 for Sonnet 4.5).

We never fill a cell with a number that no source supports. Where even an estimate isn't defensible (Windsurf Free, Cursor Free, Antigravity Free quotas), the cell says what is known, e.g. "free · unquantified", with the vendor's own words in the tooltip.

3. Included usage

Each limit is shown in the vendor's own unit: 5-hour windows, weekly pools, daily request caps, credits or $ allowances. Dollar-credit plans (Copilot, Warp, JetBrains, Zed, Augment, Amp, Kiro) state their credit amount in Included usage. Their value per $1 appears in the Real value column, after any fees or markups (Zed +10%, Augment's 40% LLM fee; Kiro credits are priced at its $0.04 add-on rate, est.). We no longer show a separate "$ credits per $1" column.

4. Best model included

This is the strongest model the plan includes without extra pay-as-you-go spending. * Anthropic's plan page names "Opus"/"Sonnet" without versions, so we assume the latest (5.5). Where versions weren't stated before, we looked them up:

Cursor's paid pools cover Grok 4.7/4.6/4.5 and Composer 2.5. Other providers' models are on-demand at API price (Premium and up).

5. Benchmarks

6. Winners

Within each tier, a solid green ★ goes to the best verified value in each column: lowest price, highest benchmark score, and highest conservative Real value multiplier. If an estimate beats every verified value, it also gets a dashed ★ est., so you see both the safe pick and the probable best. Text-only cells never win. A column needs at least two numbers and no full tie.

7. Real value

API-equivalent $ per month ÷ price, with a calculator and per-plan derivations at /value.

8. Corrections

Our monitor re-checks every source page daily, now covering 18 vendors. Measured a cap yourself? Tell us, and we'll replace estimates with data.