🔍 Read the full analysis: Inside AI Subscriptions: SemiAnalysis Finds The 5X Is A Subsidy on ThorstenMeyerAI.com
Get hardware and tech essentials 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 measured usage limits across AI subscriptions and estimated that Claude plans deliver about 5.4 to 5.6 times the API-priced usage of comparable ChatGPT plans on a coding-agent workload. The report also says OpenAI recently cut limits on its $200 plan and estimates that heavy use of premium models can make subscriptions costly to serve.
SemiAnalysis has published a comparison of usage limits across major AI subscriptions, estimating that Claude plans provide roughly 5.4 to 5.6 times the API-priced usage of similarly priced ChatGPT plans for a coding-agent workload. The analysis also tracks recent changes to OpenAI and Anthropic allowances and estimates that heavy use of premium models can weigh on providers’ inference economics.
The report says its researchers measured how subscription usage bars moved across different token types, then compared the resulting allowances with the providers’ first-party API list prices. Its central comparison uses an agentic coding workload that is mostly cached input: SemiAnalysis describes it as roughly 96.6% cached input, 2.6% cache writes, 0.4% fresh input and 0.3% output. These figures describe that workload, not a typical user’s activity.
For that workload, SemiAnalysis estimates that a $20 Claude Pro plan corresponds to about $1,178 in API usage for Claude Opus 5.5, compared with about $211 for ChatGPT Plus using GPT-6.1 Sol. At $100, the report puts Claude Max 5x at $5,725 and ChatGPT Pro 100 at $1,055. At $200, it estimates $11,726 for Claude Max 20x and $2,084 for ChatGPT Pro 200. These are modeled API-equivalent values at list prices, not cash paid out to subscribers or a guarantee of usage available in every situation.
SemiAnalysis says OpenAI recently cut token allowances on its $200 plan by roughly half. Existing subscribers keep their previous limits until October 29, while new purchases receive the reduced limits immediately, according to the report. It also describes a new $500 tier and says the tier’s headline selling point is an “Ultrafast” mode rated at 300 tokens per second; the report says it is still testing that feature.
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 Meets Inference Costs
The comparison matters because subscription limits shape both user access and the cost providers incur to serve customers. SemiAnalysis estimates that subscriptions account for about 10% of Anthropic revenue while consuming more than 40% of its inference compute. It says this mix lowers blended revenue per megawatt by roughly $36 million. These are the report’s estimates; its supplied summary does not provide the underlying calculations or a full methodology for independently checking them.
The report’s margin estimates vary sharply with usage and model choice. Assuming a subscriber uses the full allowance and an API gross margin of 92%, SemiAnalysis estimates a gross margin of about minus 369% for Opus 5.5 on the relevant plan, versus about 1% for Fable 5.1. At 20% average utilization, it estimates about 6% for Opus and 80% for Fable. Those scenarios illustrate why a plan’s nominal value does not translate directly into provider costs for every customer: actual usage and the chosen model matter.
For customers, the practical implication is that advertised prices and plan labels may not capture the full value of an allowance. Changes to model prices, usage limits, time windows and access rules can alter how much a subscriber can use. For providers, the report frames generous limits as a potential acquisition expense whose cost could change as models and pricing evolve.
As an affiliate, we earn on qualifying purchases.
Recent Pricing Changes Shift Allowances
SemiAnalysis says both companies have reduced some API prices, while subscription allowances have not always risen enough to preserve the same API-equivalent value. It reports that Fable 5.1 cut cache-read prices by 75% against Fable 5 without raising token limits. For Opus 5.5, it says input and output prices fell 20% and cache-read prices fell 60%; allowances rose about 20% on Max and 50% on Pro, which the report says did not fully offset those price reductions.
For OpenAI, the report says GPT-6.1 Sol launched without a corresponding limit increase and that the API-equivalent value of the $200 plan fell roughly 30% after the model’s cached-input price dropped. It also says the recent allowance cuts flattened the value-per-dollar difference among OpenAI’s Pro 100, Pro 200 and Pro 500 tiers. The report notes one practical distinction: OpenAI Pro plans do not have a five-hour usage window, which may help customers who use more of their monthly allowance in bursts.
At the most expensive model tier, SemiAnalysis finds the comparison closer. It says the $200 plans have broadly similar allowances for GPT-6 Astra and Claude Fable 5.1, while Claude subscribers can use the remaining half of their limit on Opus or Sonnet. The report argues that access to those mid-tier models drives much of the larger overall difference in its comparison.
As an affiliate, we earn on qualifying purchases.
Limits, Usage and Cost Assumptions
The comparison depends on SemiAnalysis’s measurements, the workload it selected and the API list prices it used. The supplied source material does not include the complete testing protocol, sample sizes or enough underlying data to reproduce the results. The report’s API-equivalent dollar amounts should be read as modeled comparisons for the stated workload, rather than a universal measure of what every subscriber will receive.
Actual costs depend on how much of an allowance customers use, which models they select and how token types are priced. The source material also does not establish how long current limits will remain in place, whether providers will make further changes, or how much the new $500 tier’s Ultrafast mode affects user experience. SemiAnalysis says it is still testing that mode.
As an affiliate, we earn on qualifying purchases.
Provider Limits Remain the Next Test
The next useful evidence will be further measurements of subscription limits and model access as providers adjust prices and plans. OpenAI’s reported grandfathering period for existing $200 subscribers runs through October 29; the source does not state what limits those subscribers will receive after that date. SemiAnalysis says it is testing the $500 plan’s Ultrafast mode, but provides no final result in the supplied material.
Readers comparing plans should check current terms directly with each provider, since allowances and model prices can change. AI subscriptions are usage products whose value depends on workload and actual consumption; the report’s estimates do not predict individual costs or performance.
As an affiliate, we earn on qualifying purchases.
Key Questions
What did SemiAnalysis compare?
It measured usage limits across AI subscriptions and estimated the API list-price value of those limits for selected models and workloads.
Where does the 5.4 to 5.6 times figure come from?
It is the report’s estimate comparing Claude and ChatGPT subscription allowances on an agentic coding workload, priced at first-party API list rates.
Did OpenAI cut the limits on its $200 plan?
SemiAnalysis says the plan’s token allowances were roughly halved. The report says existing subscribers retain previous limits until October 29, while new purchases receive the lower limits immediately.
Does the report say every Claude subscriber costs Anthropic money?
No. Its margin estimates depend on the model used and the share of the allowance consumed. SemiAnalysis estimates materially different margins for Opus 5.5 and Fable 5.1, and for full versus 20% usage scenarios.
Are these estimates a guarantee of subscription value?
No. They are modeled comparisons based on a particular workload, measured limits and API list prices. Actual access and provider costs depend on usage, model choice and current plan terms.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
