Research note 009 / AI subscriptions

Claude vs Codex: tokens per dollar by plan tier

Per week, Claude Pro supports more coding tokens than ChatGPT Plus: 71M against 52M. At $100 and $200 Codex comes out ahead, because Max 20x measured about 10× Pro per week, not 20×, while OpenAI puts Pro $200 at 20× Plus. Counting October’s 11 Codex resets, Codex gave more tokens at every tier.

How many tokens a plan can support

Figure 1

Per week, Claude leads at $20; Codex leads at $100 and $200

Coding tokens one plan supports per week, log scale. Claude on Opus 5.5, Codex on GPT-6.1 Sol.

50M100M200M500M1B2B$20 a month$100 a month$200 a monthClaude Promeasured, 2 runs71MChatGPT Plusmeasured, 3 runs52MClaude Max 5xAnthropic 5× Pro, not measured355MChatGPT Pro $100OpenAI 10× Plus, not measuredbefore 29 Oct: 262M523MClaude Max 20xmeasured 29 Oct655MChatGPT Pro $200OpenAI 20× Plus, not measuredbefore 29 Oct: 523M1.05B50M100M200M500M1B2B$20 a month$100 a month$200 a monthClaude Promeasured, 2 runs71MChatGPT Plusmeasured, 3 runs52MClaude Max 5xAnthropic 5× Pro, not measured355MChatGPT Pro $100OpenAI 10× Plus, not measuredbefore 29 Oct: 262M523MClaude Max 20xmeasured 29 Oct655MChatGPT Pro $200OpenAI 20× Plus, not measuredbefore 29 Oct: 523M1.05B
tokenusage.dev quota runs, 12 to 29 Oct 2026 · ChatGPT Pro: OpenAI multiples as reported 29 Oct
Sources: tokenusage.dev quota runs (12 to 29 Oct 2026); Pro multiples as reported by Startup Fortune. The Pro $200 “before 29 Oct” mark applies to plans bought after 29 September.

Claude Max 20x gives about 10× Pro per week, not 20×. Per week, Claude Pro gives up to 1.4× what ChatGPT Plus gives. At $200 the order flips: OpenAI’s 20× Plus for Pro $200 is about 1.6× what Max 20x measured.

With the resets

Figure 2

Counting the 11 October resets, Codex gave more coding tokens at every tier

Cumulative coding tokens for someone who uses up every allowance before the next refill¹, 1 to 30 October 2026

  • Claude
  • Codex
  • Codex reset
  • Claude Reset for free
$20 a month0350M700M630M ChatGPT Plus430M Claude Pro$100 a month02B4BOpenAI doubles Pro $100 limits,29 Oct3.4B ChatGPT Pro $1002.1B Claude Max 5x$200 a month010B20BOct 1Oct 8Oct 15Oct 22Oct 2913B ChatGPT Pro $2003.9B Claude Max 20x$20 a month0350M700M630M ChatGPT Plus430M Claude Pro$100 a month02B4BOpenAI doubles Pro $100 limits, 29 Oct3.4B ChatGPT Pro $1002.1B Claude Max 5x$200 a month010B20BOct 1Oct 8Oct 15Oct 22Oct 2913B ChatGPT Pro $2003.9B Claude Max 20x
tokenusage.dev quota runs and reset tracker (Codex resets from Tibo Sottiaux’s posts), 1 to 30 Oct 2026
Data: tokenusage.dev reset tracker; Pro multiples as reported by Startup Fortune; Claude reset rules: Anthropic.

Codex refilled 12 times in October: the start of the month plus 11 resets. Claude refilled 6 times: five weekly resets plus the one Reset for free. Each Codex reset also restarts the 7-day week, so resets add a full allowance each time.

Get alerted on the next Codex reset: tokenusage.dev can text or email you when OpenAI resets everyone’s limits.

Don’t miss the free credits

Beyond tokens, the two sides now match on API credits ($100/$200/$500): Max has carried them since 7 October, and ChatGPT Pro added them on 28 October. Our guides show how to claim them: Claude Max API credits and ChatGPT Pro API credits.

Figure 3

API credits now match; Codex adds resets and no five-hour limit on Pro

Plan benefits beyond tokens, as of 30 October 2026. Struck through: not included, or expired.

Claude Pro

$20 /month
  • Pay yearly and it’s $17 a month
  • Five-hour limit
  • Not included: Fable models inside your limits
  • Not included: Priority access at busy times
  • Not included: Monthly API credits
  • No longer offered: 50% more Claude Code weekly usage (expired Sep 13)
  • No longer offered: One Reset for free (expired Oct 22)

ChatGPT Plus

$20 /month
  • Five-hour limit
  • 11 usage resets in October
  • Not included: Dot
  • Not included: Ultrafast mode for GPT-6.1 Sol and GPT-6 Astra
  • Not included: Monthly API credits

Claude Max 5x

$100 /month
  • Five-hour limit
  • Fable models inside your limits (up to 50% of the week)
  • Priority access at busy times
  • $100 a month in API credits
  • No longer offered: 50% more Claude Code weekly usage (expired Sep 13)
  • No longer offered: One Reset for free (expired Oct 22)

ChatGPT Pro $100

$100 /month
  • No five-hour limit
  • 11 usage resets in October
  • Dot
  • $100 a month in API credits (from 28 Oct)
  • Not included: Ultrafast mode for GPT-6.1 Sol and GPT-6 Astra

Claude Max 20x

$200 /month
  • Five-hour limit
  • Fable models inside your limits (up to 50% of the week)
  • Priority access at busy times
  • $200 a month in API credits
  • No longer offered: 50% more Claude Code weekly usage (expired Sep 13)
  • No longer offered: One Reset for free (expired Oct 22)

ChatGPT Pro $200

$200 /month
  • No five-hour limit
  • 11 usage resets in October
  • Dot
  • $200 a month in API credits (from 28 Oct)
  • Not included: Ultrafast mode for GPT-6.1 Sol and GPT-6 Astra
  • No longer offered: GPT-5.3-Codex-Spark, with limits of its own (expired Sep 14)
API credits don’t raise subscription limits, and Codex’s don’t cover Codex run on an API key.
Sources: Claude plan cards; ChatGPT plan cards; Claude Code promotion; Spark retirement; Reset offer expiry.

Both vendors’ API credits expire at the end of each billing cycle and don’t raise subscription limits. Sources: Anthropic: What is a limit reset?, Anthropic: Claude Code weekly-limits promotion, OpenAI: Codex pricing, OpenAI: API credits for ChatGPT subscribers, OpenAI API changelog, Oct 28, 2026, tokenusage.dev reset tracker, and the plan cards on Claude Pro vs Max and Codex pricing.

How we measured

We ran the same ten-turn coding sessions on each plan and read its weekly usage meter before and after. A plan’s weekly figure is the run’s tokens divided by the share of the week it used. Claude Pro is the average of two runs (12 and 26 October), and ChatGPT Plus the median of three (12, 20 and 26 October), both on their flagship models, Opus 5.5 and GPT-6.1 Sol. Max 20x is one Opus 5.5 run on 29 October.

The other plans are estimates, not measurements: Max 5x uses Anthropic’s 5× Pro, and the ChatGPT Pro plans use OpenAI’s multiples of Plus as reported by Startup Fortune. The meters show whole percents, so each measured figure is approximate. Codex resets come from our reset tracker of Tibo Sottiaux’s posts.

A token is not the same work on every model. At API list prices, a Claude Pro coding week from our 12 October run is worth about $227 of Opus 5.5, and a ChatGPT Plus week from its 12 October run about $47 of GPT-6.1 Sol.

¹ Upper bound. It assumes someone uses up every allowance before the next refill. Most Codex resets arrived unannounced, Plus needs about 30 hours of five-hour windows to use up a week, and a light user gains nothing from a reset. Both vendors start October with a fresh allowance, and Claude’s weekly refills are counted from 1 October.