tokens-per-dollar · prices read

Claude Max vs API: what your usage costs

Claude Max costs $100 a month (Max 5x) or $200 (Max 20x). The API has no fee and bills every token: on Claude Opus 5.5, $4.00 per million input tokens and $20.00 per million output. Max is the cheaper of the two only when your month would cost more than its price at API rates and its allowance holds your busiest 5 hours and your busiest week. Anthropic publishes that allowance only as 5 or 20 times Pro's per-session allowance, never in tokens or dollars, so no calculator can tell you which is cheaper for you. Your own logs can: npx --allow-git=root github:agentwares/tokens-per-dollar prices your last 30 days at API rates and shows what the allowance would have to hold.

What Anthropic publishes

PlanPer month, billed monthlyBilled annuallyAllowance
Claude Pro$20$17Pro: the unit the others are stated in
Claude Max 5x$100monthly billing only5× Pro per session
Claude Max 20x$200monthly billing only20× Pro per session
Claude Team standard$25 a seat, 2 seats minimum$201.25× Pro per session
Claude Team premium$125 a seat, 2 seats minimum$1006.25× Pro per session

Every plan has a 5-hour session allowance and a weekly limit across all models, shared with Claude on the web and desktop. None is published as a number of messages, hours or tokens. Past the allowance, extra usage is billed at standard API rates. On Pro and Team standard, Fable models run on usage credits only, that is at API rates from the first token; on Max and Team premium they can use up to half of the weekly limit.

What a dollar buys on the API

ModelInputCache readsOutput
Claude Fable 5.1100,000 tokens4,000,000 tokens20,000 tokens
Claude Opus 5.5250,000 tokens5,000,000 tokens50,000 tokens
Claude Sonnet 5.5500,000 tokens5,000,000 tokens100,000 tokens
Claude Haiku 4.51,000,000 tokens10,000,000 tokens200,000 tokens

Tokens per US dollar at Anthropic's list prices (read 7 October 2026). A coding agent sends its whole context again on every request, and most of it is read from the cache, so the cache-read price matters as much as input and output. At these prices, Max 5x's $100 is 5 million Opus 5.5 output tokens, or 500 million cache reads.

What it looks like on a month of use

The Claude Code part of npx --allow-git=root github:agentwares/tokens-per-dollar, on synthetic history. Each Claude plan shows its fee as “from”, because the allowance that decides whether the fee is all you pay is not published; the line under the table gives the busiest 5 hours and week it would have to hold. Copilot rows are an estimate: Copilot keeps no local logs, so they are the same tokens at GitHub's per-token rates.

$ npx --allow-git=root github:agentwares/tokens-per-dollar demo
Claude Code: 2,313 requests, 170.0M tokens, $241.93 at Anthropic API list prices
  busiest 5 hours $29.82 · busiest week $109.21 (API list prices)

  plan                  fee/mo  a month like this            how
  Copilot Max             $100            $117.28  estimate  $217.28 of credits, $200.00 included ²
  Copilot Pro+             $39            $186.28  estimate  $217.28 of credits, $70.00 included ²
  Copilot Pro              $10            $212.28  estimate  $217.28 of credits, $15.00 included ²
  Anthropic API              —            $241.93  priced    list prices × your tokens
  Claude Pro               $20        from $20.00            allowance not published ¹
  Claude Team standard     $25        from $25.00            1.25× Pro a session · 2 seats min ¹
  Claude Max 5x           $100       from $100.00            5× Pro a session ¹
  Claude Team premium     $125       from $125.00            6.25× Pro a session · 2 seats min ¹
  Claude Max 20x          $200       from $200.00            20× Pro a session ¹

  Cheapest priced: Copilot Max, $117.28 a month (estimate), $124.65 less than Anthropic API
    ($241.93).
  Could be cheaper, if the allowance Anthropic does not publish holds your busiest 5 hours ($29.82)
    and week ($109.21): Claude Pro from $20.00, Claude Team standard from $25.00, Claude Max 5x from
    $100.00.
Synthetic history: the real output of npx --allow-git=root github:agentwares/tokens-per-dollar demo on generated transcripts, not anyone's usage (as of 7 October 2026).

Changelog: what the vendors changed, and when

Dates are the vendors' own. “As reported” means the press quoting the vendor, because the vendor's own post could not be read. Each entry says what changed, never why.

Run it on your own logs

On the machine where you use Claude Code or Codex:

$ npx --allow-git=root github:agentwares/tokens-per-dollar

It reads the transcripts Claude Code and Codex keep on your machine, takes the token counts, and prices your last 30 days on every plan on this page. Free and MIT; it makes no network request and sends nothing. --json prints every price with its source and the date it was read.

For a team: seatledger keeps everyone's coding-agent ledger in one place and flags when the token rate moves. Free for up to 3 developers; email alerts on Team, $49 a month.

Sources