🔍 Read the full analysis: What You’re Paying For In AI Subscriptions: The 5X Subsidy on ThorstenMeyerAI.com
Get office and shipping supplies delivered free — and shop member deals
- Fast, free delivery on millions of items
- Access to Prime Big Deal Days deals on October 6–7
- Prime Video, Amazon Music and more included
TL;DR
SemiAnalysis compared usage limits across major AI subscriptions by measuring how token use moves each service’s meter, then pricing that usage at API list rates. In its agentic workload comparison, Claude tiers returned about 5.4 to 5.6 times the API-priced value of comparable ChatGPT tiers, but the estimate depends on the workload and current limits. Recent price and allowance changes on both services show why subscription value can shift without a change in the monthly fee.
SemiAnalysis has published a comparison of usage limits across major AI subscriptions, finding that Claude plans provide about 5.4 to 5.6 times the API-priced value of comparable ChatGPT plans on a tested agentic workload. The estimate translates measured subscription usage into the cost of the same tokens at API list prices; it is not a cash value or a guarantee that every subscriber can use the full allowance.
The comparison covers subscriptions from OpenAI, Anthropic, Meta, SpaceXAI, Cursor, Cognition, Z.ai, MiniMax and Moonshot. SemiAnalysis says it tested how different token types affected each provider’s usage meter, then calculated the API-equivalent cost of the measured limits. Its headline side-by-side comparison uses GPT-6.1 Sol and Claude Opus 5.5 on a coding-agent workload made up mostly of cached input: about 96.6% cached input, with roughly 0.4% fresh input, 2.6% cache writes and 0.3% output.
At $20 per month, the report estimates $211 in API-priced usage for ChatGPT Plus and $1,178 for Claude Pro. At $100, it estimates $1,055 for ChatGPT Pro 100 and $5,725 for Claude Max 5x. At $200, its figures are $2,084 for ChatGPT Pro 200 and $11,726 for Claude Max 20x. These are the report’s calculations using each model’s list prices and the tested workload. The report says the gap remains large when comparing raw token counts, although the more expensive Opus tokens increase its dollar-valued result.
The same report describes a recent change to OpenAI’s $200 plan: its token allowances were roughly halved, according to SemiAnalysis’s tracking. Existing subscribers keep their previous limits until 29 October; new purchases receive the lower limits immediately. OpenAI also introduced a $500 tier, whose main advertised distinction is an “Ultrafast” mode of 300 tokens per second. SemiAnalysis says it is still testing that mode and estimates the new tier provides only about 21% more Astra usage than the old $200 plan.
The 5x is a subsidy, not a price
SemiAnalysis metered the meters — every major AI subscription, token type by token type, converted to API list value. On the mid-tier models both labs call the daily driver, a Claude plan returns ~5–6× the API value of the matching ChatGPT plan. Real — and the least durable number in the report.
…and the plan is fully exhausted. One pool for every model.
…and the plan is only half used — Fable is capped at 50% of the limit, leaving the rest for Opus/Sonnet. That’s where the mid-tier gap compounds.
- $200 plan halved — Sol-class value down >50% (6.1 Sol cache price cut compounds it)
- Old limits kept until 29 October; new buyers cut immediately
- New $500 tier: only +21% Astra vs the old $200 — real draw is 300 TPS Ultrafast
- Ladder flattened: Pro 100/200/500 now identical per dollar; multipliers removed from pricing page
- In OpenAI’s favour: no 5-hour window on Pro plans — easier to use the full allowance
- Flat per-dollar value across all tiers, before and after
- New premium models placed at lower relative limits (Fable capped at 50%)
- Opus allowances raised ~20% (Max) / ~50% (Pro) with the 5.5 price cut — not enough to fully offset it
- Repeatedly walked back planned cuts earlier this year under pressure from OpenAI’s generosity
- Twelve months ago, OpenAI was the generous option. Positions swap.
Gross margin per plan, assuming 92% API gross margins. The subsidy lives almost entirely in Opus and Sonnet usage — Anthropic would already be near software-like subscription margins if everyone used only Fable. Subscriptions matter even more for OpenAI, where they’re a larger share of revenue.
Three identical subscriptions; one had ~20% lower limits. The provider (unnamed) confirmed an “extremely tiny” A/B test on limit balancing. Two lessons: limits can change silently, per account, at any time — and you won’t know without instrumentation. The usage bar is a percentage, not a contract.
If you’re choosing a plan this month for agentic coding on a mid-tier model, the report settles it: a Claude plan returns ~5–6× the API value of the matching ChatGPT plan. But a plan returning 58× its fee on a model served at a steeply negative margin for heavy users is a marketing budget with a usage meter. Value moves silently, gets A/B tested per account, and twelve months ago ran the other way. Use the subsidy while it exists — it’s genuinely large. Don’t build a cost model on it. Price workloads at API rates, keep a router between you and any one vendor, and benchmark open weights on your own hardware for steady volume. A deal you can’t verify isn’t a price. It’s weather.
Subscription Value Depends on Usage
The estimates show why a monthly price alone does not reveal what a subscriber receives. A plan’s practical value depends on which models a customer uses, how much of the allowance they can consume, and how the provider defines its limits. SemiAnalysis says OpenAI plans do not impose a five-hour usage window, which may help people who need to spend a larger share of their monthly allowance in bursts. That difference can matter even when API-equivalent totals favor Claude in the report’s chosen workload.
The report also links generous limits to providers’ serving costs. SemiAnalysis estimates subscriptions account for about 10% of Anthropic revenue but can use more than 40% of its inference compute, reducing blended revenue per megawatt by roughly $36 million. These are the report’s estimates, not audited company disclosures. Its modeled margins vary sharply with utilization: at full use, it estimates a deeply negative gross margin for Opus 5.5 subscriptions and about 1% for Fable 5.1; at 20% average use, the estimates rise to about 6% and 80%, respectively.
Those calculations help explain why providers may alter limits as model and serving costs change. They also put the 5× figure in perspective: it describes a specific comparison at a point in time, while the underlying allowance and API prices can move. Subscribers comparing plans should read model-specific limits and time windows alongside the advertised monthly fee.
As an affiliate, we earn on qualifying purchases.
Price Cuts and Limits Shift
The report describes price changes at both companies. Anthropic reduced cache-read prices for Fable 5.1 by 75% compared with Fable 5, and cut Opus 5.5 input and output prices by 20% and cache reads by 60% compared with Opus 5. SemiAnalysis says Fable 5.1 launched without a corresponding increase in token limits. It estimates Opus allowances rose about 20% on Max and 50% on Pro, increases that did not fully offset the lower API prices in its value calculation.
OpenAI’s GPT-6.1 Sol also had a lower cached-input price without a corresponding allowance increase, according to the report. SemiAnalysis estimates this reduced API-equivalent value on the $200 plan by about 30%. The report says OpenAI’s Pro tiers now return similar tokens per dollar, whereas the previous ladder offered progressively greater value at higher tiers. OpenAI also removed “5x more usage” and “20x more usage” comparisons from its pricing page, the report says.
For frontier models, the comparison is closer. SemiAnalysis estimates a $200 ChatGPT plan would be exhausted after about $2,897 of GPT-6 Astra usage at API prices. It says Claude’s Fable 5.1 would use about half of the Claude plan’s limit at an equivalent $2,485, with the remaining allowance available for Opus or Sonnet. That model mix is one reason the report’s largest gap appears in its mid-tier comparison.
Limits and Costs Still Move
The 5.4–5.6× estimate is tied to the report’s chosen models, token mix, API list prices and measured limits. Actual subscribers may use different models or token types, and may not consume their full allowance. The comparison does not establish how much any individual customer saves against buying API usage directly.
Some details are also unsettled. SemiAnalysis says it is still testing OpenAI’s 300-token-per-second Ultrafast mode. The source material does not provide a method for independently auditing the companies’ total subscription compute use or the report’s margin estimates. It is also unclear how providers will adjust limits as they change model prices and capacity.
Watch the Next Limit Changes
The immediate milestone for existing ChatGPT Pro 200 subscribers is 29 October, when SemiAnalysis says their grandfathered limits end. The report’s next open question is whether OpenAI’s Ultrafast mode delivers enough practical benefit to distinguish its $500 tier from lower plans.
For both providers, future changes to API prices, subscription allowances and model access will alter the comparison. SemiAnalysis’s figures offer a snapshot of measured limits and current list prices; readers will need updated measurements to know whether the estimated gap persists.
Key Questions
What does the reported 5–6× value mean?
It is SemiAnalysis’s estimate of the API list-price cost of the usage allowed by comparable plans on its tested agentic workload. It does not mean subscribers receive cash worth five or six times their fee.
Which plans were compared?
The report compared ChatGPT and Claude tiers at $20, $100 and $200 per month, using GPT-6.1 Sol and Claude Opus 5.5 for its central agentic-workload calculation.
Did OpenAI change its $200 plan?
According to SemiAnalysis, OpenAI roughly halved the plan’s token allowances. Existing subscribers keep their old limits until 29 October, while new purchases receive the reduced limits immediately.
Does the estimate apply to every subscriber?
No. It depends on the report’s workload, model mix, token types, current limits and API prices. A subscriber’s actual usage and value may differ.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
