🔍 Read the full analysis: The 5X Is A Subsidy: A SemiAnalysis Look Inside AI Subscriptions on ThorstenMeyerAI.com
Get monitors, keyboards and dev gear 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 major AI subscriptions and converted them into equivalent API list-price value. Its comparison found Claude’s mid-tier plans offered about 5 to 6 times the API-equivalent value of comparable ChatGPT plans for a specified coding-agent workload, while estimating that heavy use can make subscriptions costly to serve. The calculations depend on the tested workload, model prices and usage assumptions, and do not establish each provider’s actual per-subscriber profitability.
SemiAnalysis has published a cross-provider comparison of AI subscription usage limits, estimating how much the included usage would cost at each provider’s API list prices. For a specified coding-agent workload, the report says Claude’s mid-tier plans delivered about five to six times the API-equivalent value of comparable ChatGPT plans, while warning that heavy use can make those subscriptions expensive to serve.
The report tested usage across subscriptions from Anthropic, OpenAI, Meta, SpaceXAI, Cursor, Cognition, Z.ai, MiniMax and Moonshot. It tracked how provider usage bars changed across token types, then translated the measured allowances into estimated API prices. The comparison is not a direct measure of providers’ costs or profits: it uses public list prices and a defined workload.
For that workload, SemiAnalysis compared Claude Opus 5.5 with GPT-6.1 Sol. It described the coding-agent traffic as heavily cache-weighted: about 96.6% cached input, 2.6% cache writes, 0.4% fresh input and 0.3% output. At $20 a month, the report put Claude Pro’s full allowance at roughly $1,178 in API-equivalent value, against about $211 for ChatGPT Plus. At the $100 and $200 tiers, its reported Claude-to-ChatGPT ratios were about 5.4 to 5.6 times.
SemiAnalysis said both companies’ plans returned roughly similar API-equivalent value per subscription dollar within their own tiers: about 10.5 times the fee for the OpenAI plans measured and about 58 times for Anthropic’s. Those figures are specific to the chosen models, prices and workload. The report said the gap remains large when counted in raw tokens, although the cheaper per-token price of GPT-6.1 Sol makes a dollar-based comparison less direct.
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.
The Cost of Heavy Subscription Use
The report’s larger point is that subscription generosity has an inference-cost trade-off. SemiAnalysis estimates subscriptions account for about 10% of Anthropic revenue but can consume more than 40% of its inference compute. It estimates this mix lowers blended revenue per megawatt by roughly $36 million. These are the firm’s estimates, not company-reported financial results, and the supplied material gives limited detail on their underlying assumptions.
SemiAnalysis modeled margins under different levels of use. Assuming a subscriber fully uses the allowance and API gross margins of 92%, the report calculated an estimated gross margin of about minus 369% for a plan used entirely on Opus 5.5, compared with about 1% for one used on Fable 5.1. At 20% average utilization, its estimates rose to about 6% and 80%, respectively. These are scenarios, not observed subscriber-level margins; actual results would depend on usage patterns, serving costs and the mix of models.
The analysis gives customers a way to compare plans beyond their monthly sticker prices, but the headline multiple is not a promise that every subscriber receives that value. It also highlights why providers may change limits as model prices fall: lower API prices can reduce the calculated value of a fixed subscription allowance, unless providers add more usage.
As an affiliate, we earn on qualifying purchases.
Recent Changes to Plan Limits
The comparison reflects changes SemiAnalysis says OpenAI made the previous week. The firm reports that OpenAI halved usage allowances on its $200 plan, with the reduction applying immediately to new purchases. Existing subscribers, it says, keep prior limits until October 29. OpenAI also introduced a $500 tier. SemiAnalysis estimates that this tier provides about 21% more Astra usage than the former $200 plan, while offering less Sol-class API-equivalent value after a price cut to cached input.
According to the report, OpenAI’s Pro 100, Pro 200 and Pro 500 plans now provide similar tokens per dollar, flattening the previous ladder of progressively higher value. OpenAI also removed “5x more usage” and “20x more usage” comparisons from its pricing page. The report identifies one practical distinction: its measured OpenAI Pro plans do not have a five-hour usage window, which may help users who need to concentrate usage into a short period. It says the new $500 plan’s advertised 300-tokens-per-second “Ultrafast” mode was still being tested.
SemiAnalysis also describes price and allowance changes at Anthropic. It says Fable 5.1’s cache-read price fell 75% from Fable 5, with no corresponding increase in its token limits. Opus 5.5 prices fell for input, output and cache reads, while allowances rose by about 20% on Max and 50% on Pro. The firm says those increases did not fully offset the price reductions in its API-equivalent calculations. It reports a similar effect for OpenAI’s Sol: lower cached-input prices without an allowance change reduced the $200 plan’s calculated value by about 30%.
As an affiliate, we earn on qualifying purchases.
Limits of the Value Estimates
The comparison depends on a particular workload, including its unusually high share of cached input, and on API list prices that may change. The source material does not provide a full testing protocol, sample size, or independent verification of every measured usage limit. Results for ordinary chat, other coding tasks or different model mixes may differ.
The estimates also do not show how much a typical subscriber uses, what each provider pays to serve that usage, or how subscription revenue and compute costs are allocated. SemiAnalysis’s margin figures are modeled scenarios, not audited results. The supplied material does not include responses from OpenAI or Anthropic about the measurements or their subscription economics. It is also unclear whether the reported limits and prices have changed since the tests were conducted.
As an affiliate, we earn on qualifying purchases.
Further Testing and Plan Changes
SemiAnalysis said it was still testing OpenAI’s Ultrafast mode, so its performance and practical value were not established in the material provided. Further measurements could also show whether provider limits remain stable as models and API prices change. Subscribers should check current plan terms rather than assume the allowances in the report still apply.
The next useful evidence would include updated usage tests across different workloads, clearer information on how providers set and revise limits, and company disclosures that allow subscription revenue and inference costs to be compared directly. Until then, the report offers a snapshot of measured plan value and modeled economics, not a definitive account of either company’s subscription profitability.
As an affiliate, we earn on qualifying purchases.
Key Questions
What did SemiAnalysis compare?
It measured usage allowances across a range of AI subscriptions and estimated what the included usage would cost at each provider’s API list prices. Its headline comparison focused on Claude Opus 5.5 and GPT-6.1 Sol for a cache-heavy coding-agent workload.
Does the report prove Claude plans are five times more profitable for users?
No. The reported five-to-six-times figure is an API-equivalent value estimate for the tested workload and prices. It does not establish how much value a typical user receives or compare providers’ actual costs.
What does the report say changed in OpenAI’s plans?
SemiAnalysis says OpenAI halved allowances on its $200 plan, introduced a $500 tier and brought the measured Pro plans to similar tokens per dollar. It says existing $200 subscribers retain their previous limits until October 29, while new purchases receive the lower limits immediately.
Why can a lower API price reduce subscription value?
The report values subscription usage by pricing the included tokens at API list rates. If an API price falls but the subscription allowance stays the same, the calculated API-equivalent value also falls.
Are the reported subscription margins actual company results?
No. They are SemiAnalysis’s modeled scenarios based on assumed utilization and a 92% API gross margin. The source material does not provide audited margins for individual subscription plans.
Source: ThorstenMeyerAI.com
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
