🔍 Read the full analysis: The 5X Is A Subsidy, Not A Price: What SemiAnalysis Found Inside AI Subscriptions on ThorstenMeyerAI.com
Get the latest gadgets 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 estimating what the included tokens would cost at providers’ API list prices. Its analysis found Claude plans delivered about 5.4 to 5.6 times the API-equivalent value of comparable ChatGPT plans on the tested mid-tier models, while also documenting recent changes to OpenAI and Anthropic limits. The figures depend on workload, model prices and how much subscribers use; SemiAnalysis estimates subscriptions can consume far more inference compute than their share of revenue suggests.
SemiAnalysis has measured usage limits across major AI subscriptions and compared the included tokens with what equivalent usage would cost at API list prices. In its agentic-workload comparison, Claude plans returned about 5.4 to 5.6 times the API-equivalent value of similarly priced ChatGPT plans on the tested mid-tier models, a gap the report says remains substantial even when comparing raw token counts.
The analysis covered subscriptions from Anthropic and OpenAI, as well as plans from Meta, SpaceXAI, Cursor, Cognition, Z.ai, MiniMax and Moonshot. SemiAnalysis says it tested usage by token type, tracking how each service’s usage meter moved per million tokens. It then priced the estimated plan allowances at first-party API list rates. The comparison is an estimate of the value of the full monthly allowance, not a measure of what every subscriber actually consumes.
For the tested agentic workload, SemiAnalysis reported API-equivalent values of $211 for ChatGPT Plus and $1,178 for Claude Pro, both priced at $20 per month. At $100, ChatGPT Pro 100 came to $1,055 and Claude Max 5x to $5,725. At $200, ChatGPT Pro 200 came to $2,084, compared with $11,726 for Claude Max 20x. The report says the workload was dominated by cached input: about 96.6% of tokens, with roughly 0.4% fresh input, 2.6% cache writes and 0.3% output.
SemiAnalysis also describes recent limit changes at both companies. It says OpenAI roughly halved allowances for its $200 plan, with new purchases receiving the reduced limits immediately and existing subscribers retaining their prior limits until 29 October. The company also introduced a $500 tier. Separately, the report says Anthropic increased Opus allowances by about 20% on Max and 50% on Pro, while Fable 5.1 launched without increased token limits. These figures and descriptions are the report’s measurements and analysis.
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 Compute Costs
The comparison matters because the advertised monthly fee does not show how much compute a heavy user may consume. SemiAnalysis estimates subscriptions account for about 10% of Anthropic revenue while using more than 40% of its inference compute. The report says this can lower blended revenue per megawatt by about $36 million. It describes subscriptions as a way to attract customers, while also showing how costly high usage can be.
Its margin estimates illustrate how strongly the result depends on usage and model choice. Assuming a subscriber uses the full allowance and API gross margins are 92%, SemiAnalysis estimates that maxing out Opus 5.5 would produce a gross margin of about minus 369% on the plan. For Fable 5.1, it estimates about 1%. At 20% average utilization, the corresponding estimates rise to about 6% and 80%. These are modeled estimates, not reported company financial results.
For subscribers, the practical value can shift without a change to the monthly fee. API price cuts can reduce the API-equivalent value of an allowance if token limits do not rise enough to offset them. SemiAnalysis says OpenAI did not change Sol-class limits when GPT-6.1 shipped, and estimates API-equivalent value on the $200 plan fell by about 30% after a cached-input price cut. It also reports that Anthropic’s model price reductions were accompanied by different allowance changes across Fable and Opus.
As an affiliate, we earn on qualifying purchases.
How the Plans Were Compared
SemiAnalysis’s core measure is the estimated cost of buying a subscription’s full monthly token allowance at the provider’s public API list rates. That lets it compare plans with different prices and usage limits, but the result depends on which model is used, the mix of cached and fresh input, output volume and other workload characteristics. The featured comparison uses an agentic coding workload with a particularly large share of cached input.
The report says frontier-tier limits are closer than the mid-tier comparison suggests. On a $200 plan, it estimates OpenAI’s GPT-6 Astra allowance would be exhausted after about $2,897 of API-priced usage. Anthropic’s Claude Fable 5.1 would use about half of its plan’s limit after $2,485 of API-priced usage, because Fable can consume only 50% of the Claude plan allowance; the rest remains available for Opus or Sonnet.
SemiAnalysis says OpenAI’s earlier plan ladder offered increasing value per dollar at higher tiers, but that following the change, its Pro 100, 200 and 500 plans returned the same number of tokens per dollar in the report’s comparison. It also notes that OpenAI Pro plans have no five-hour usage window, which may help users who need to spend more of their monthly allowance in a short burst.
Limits of the Value Estimates
The API-equivalent figures do not show each subscriber’s actual usage, the companies’ realized serving costs or the profitability of individual plans. SemiAnalysis’s margin calculations rely on assumptions, including a 92% API gross margin and specified utilization rates. The report’s headline ratios also reflect a particular agentic workload, with a large proportion of cached input; a different mix of tokens or models could change the result.
It is not clear from the supplied report material how long the measured limits will remain in place, whether companies will make further changes or what average utilization looks like across their full subscriber bases. SemiAnalysis says it is still testing the $500 plan’s 300-tokens-per-second “Ultrafast” mode, so its practical performance and value are not established here.
Further Testing and Plan Changes
SemiAnalysis says it is still testing OpenAI’s Ultrafast mode on the new $500 tier. Further results could clarify whether its advertised speed changes the plan’s practical value, particularly for users who prioritize faster responses over total monthly usage.
Subscribers can watch for updates to usage limits, model pricing and plan terms. Those measures jointly determine the API-equivalent value in SemiAnalysis’s method, so the reported ratios describe the plans at the time of the comparison rather than a permanent price relationship. The report material does not identify a date for another full comparison.
Key Questions
What does the reported 5.4 to 5.6 times figure measure?
It compares the estimated API list-price value of the full monthly allowance in similarly priced Claude and ChatGPT plans on the tested agentic workload. It is not a claim that every subscriber receives or uses that amount.
Which plans were compared?
The report compared ChatGPT Plus at $20 with Claude Pro at $20, ChatGPT Pro 100 at $100 with Claude Max 5x at $100, and ChatGPT Pro 200 at $200 with Claude Max 20x at $200.
What changed in OpenAI’s $200 plan?
SemiAnalysis says the plan’s token allowances were roughly halved. It reports that new purchases received the lower limits immediately, while existing subscribers kept their previous limits until 29 October.
Does a lower API price automatically give subscribers more value?
No. Under SemiAnalysis’s method, lower API prices reduce the estimated API-equivalent value of a fixed allowance unless the provider also raises the token limits enough to offset the price cut.
How certain are the report’s margin estimates?
They are estimates based on assumptions about API gross margins, subscriber utilization and model usage. They do not establish the actual profit or loss on each company’s subscription plans.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
![Competitive Programmer's Handbook (2018) [Pdf]](https://1023jack.com/wp-content/uploads/2026/10/competitive-programmer-s-handbook-2018-pdf-featured-260x140.jpg)