BenchLeader

Claude Ultra: limits, models and real cost

Claude Ultra costs $200 a month. It runs Claude Opus 5.5, Claude Sonnet 5.5 and Claude Haiku 5.5. Anthropic describes the allowance as “10x Pro usage every week; $200/month in API credits for the Claude Platform”. How it refreshes: weekly limit only once the account is verified (no five-hour limit), plus a separate weekly Fable limit. Used in full on Claude Opus 5.5, the allowance comes to about 32 billion tokens a month by Real API Pricing's medium-confidence estimate: about $12,426 at API prices, 62× the subscription.

Price
$200
a month
Allowance in dollars
$12,426
estimate on Claude Opus 5.5
API usage per dollar
62×
at full use
Break-even
625 sessions
a month on Claude Opus 5.5

About Claude Code

Claude Code and the Claude apps draw on the same allowance, so coding competes with chat. Max and Ultra (renamed from Max 5x and Max 20x on 30 October 2026) are billed monthly only. Ultra has no five-hour limit once the account is verified, and its weekly limit is 10x Pro, not 20x. Both include monthly API credits for the Claude Platform, which sit outside the plan's usage pool.

Limits and reset windows

In Anthropic’s words“10x Pro usage every week; $200/month in API credits for the Claude Platform”
How it refreshesWeekly limit only once the account is verified (no five-hour limit), plus a separate weekly Fable limit
How the size is statedOnly as a multiple of another tier
Five-hour limitNone once verified
Weekly limitYes
Shared with chatYes
API credits$200 a month for the Anthropic platform, outside the plan’s usage pool

Models, effort and what a task costs

Artificial Analysis measured Claude Opus 5.5 at 5 efforts. One task of its Intelligence Index costs $0.98 through the API at low effort and $10.04 at max; through Claude Ultra at full use, about $0.016 and $0.16. In T3 Code's usage data an average Claude Opus 5.5 turn uses about 3.0 million tokens, 98% of them read back from cache, and costs $1.02 at API list price, so a full allowance holds roughly 12,224 such turns a month.

ModelEffortIndexOne task, APIOne task, planTasks a month
Claude Opus 5.5low58.9$0.98$0.01612,738
Claude Opus 5.5medium61.2$2.45$0.0395,070
Claude Opus 5.5high66.0$3.46$0.0563,586
Claude Opus 5.5xhigh65.4$6.05$0.0972,055
Claude Opus 5.5max68.4$10.04$0.161,238
Claude Sonnet 5.5low50.7$0.51$0.008224,301
Claude Sonnet 5.5medium50.6$0.78$0.01315,917
Claude Sonnet 5.5high59.5$1.64$0.0267,595
Claude Sonnet 5.5xhigh62.5$4.09$0.0663,037
Claude Sonnet 5.5max64.2$10.02$0.161,240
Claude Haiku 5.5low47.7$0.044$0.0010206,900
Claude Haiku 5.5medium50.1$0.092$0.002098,670
Claude Haiku 5.5high55.3$0.16$0.003557,101
Claude Haiku 5.5xhigh56.7$0.25$0.005536,455
Claude Haiku 5.5max55.3$0.41$0.009122,028

One task is an average task of Artificial Analysis’ Intelligence Index, at each effort it was measured at. Through the plan it is the API cost scaled by what the plan charges against its estimated allowance, so it holds only if you use the whole allowance. Effort changes how fast an allowance runs out, not its size.

What a real agent turn costs

A turn is one prompt run to the end, every tool call and re-read of the context included, averaged over T3 Code’s users and priced at API list price. The last column divides the allowance's API value by that price.

ModelTokens per turnFrom cacheAPI price per turnTurns a month on this plan
Claude Opus 5.53.0 million98%$1.0212,224
Claude Sonnet 5.52.6 million97%$0.5124,444
Claude Haiku 5.51.7 million96%$0.030298,275

Who it suits

  • Long unbroken sessions, once the account is verified: then there is no five-hour limit, only a weekly one.
  • Running Claude Opus 5.5, the highest-ranked model on any plan here (BenchLeader Index 68.4).
  • Building on the Anthropic API as well: $200 a month of platform credits come with it, outside the plan's usage pool.

Think twice for

  • Heavy chat use alongside coding: chat in Anthropic's apps spends the same allowance.

Every Claude Code tier

TierPriceAllowance at full usePer dollar
Claude Pro$20About $1,200 on Claude Opus 5.5 (estimate)60×
Claude Max$100About $7,224 on Claude Opus 5.5 (estimate)72×
Claude Ultra$200About $12,426 on Claude Opus 5.5 (estimate)62×

Against the closest plans

The nearest plan in price from each of three other vendors.

PlanPriceAllowance at full usePer dollarFive-hour limitBest model
Claude Ultra$200About $12,426 on Claude Opus 5.5 (estimate)62×None once verifiedClaude Opus 5.5
Codex Pro 200$200About $5,604 on GPT-6 Astra (estimate)28×NoGPT-6 Astra
Cursor Ultra$200No figure–Not statedGrok 4.6
Google AI Ultra 20x$199.99About $5,691 on Gemini 3.8 Flash (rough estimate)28×Not statedGemini 4 Argon

Head to head

What changed

  • 1 Nov 2026 — Max 5x and Max 20x renamed Max and Ultra, at the same prices. Ultra corrected to 10x Pro's weekly usage, as Anthropic's pricing page now states (it was listed here as 20x), with no five-hour limit for verified accounts since 30 October. Source
  • 30 Oct 2026 — Claude Max split into Max 5x ($100) and Max 20x ($200), with their monthly API credits and weekly limits. Source

Claude Ultra: questions people ask

How much does Claude Ultra cost?
Claude Ultra costs $200 a month. Price read from Anthropic's pricing page on 30 Oct 2026.
What are Claude Ultra's usage limits?
Anthropic describes the allowance as “10x Pro usage every week; $200/month in API credits for the Claude Platform”. How it refreshes: weekly limit only once the account is verified (no five-hour limit), plus a separate weekly Fable limit. It has a weekly limit but no five-hour limit once the account is verified. Chat in Anthropic's apps uses the same allowance.
Which models does Claude Ultra include?
Claude Opus 5.5, Claude Sonnet 5.5 and Claude Haiku 5.5. The highest-ranked is Claude Opus 5.5, at 68.4 on the BenchLeader Index. Effort changes how fast the allowance runs out, not its size: one index task on Claude Opus 5.5 costs 10.3 times as much at max as at low.
Is Claude Ultra worth it?
Used in full on Claude Opus 5.5, the allowance comes to about 32 billion tokens a month by Real API Pricing's medium-confidence estimate: about $12,426 at API prices, 62× the subscription. Against paying per token, it pays for itself after about 625 agent sessions a month on Claude Opus 5.5. Every figure assumes the whole allowance is used; an unused month makes every task dearer.
Claude Ultra or Codex Pro 200?
Both cost $200 a month. Claude Ultra goes further per dollar: at full use it is worth about 62× its price in API usage on Claude Opus 5.5, against 28× for Codex Pro 200 on GPT-6 Astra.

Sources

  • Price and the allowance wording: claude.com, read 30 Oct 2026.
  • Limits: Claude usage limits, read 30 Oct 2026.
  • Limits: Max and Ultra announcement, read 30 Oct 2026.
  • Allowance estimate: Real API Pricing (MIT licence), data to 27 Sep 2026, priced at BenchLeader’s API prices of 16 Nov 2026.
  • Cost per index task at each effort: Artificial Analysis, as of 16 Nov 2026.
  • Real agent turns: T3 Code’s usage telemetry as published by Slopalytics, priced at API list price.

Every plan side by side, with how the estimates are made, is on the coding plans page.

Cite as: BenchLeader, “Claude Ultra: limits, models and real cost”, data as of 16 Nov 2026, https://www.benchleader.com/coding-plans/claude-ultra.