🔍 Read the full analysis: Unpacking The Subsidy Behind AI Subscription Pricing on ThorstenMeyerAI.com
TL;DR
SemiAnalysis compared the usage limits of major AI subscriptions with the cost of buying equivalent tokens at API list prices. In its coding workload comparison, Claude’s mid-tier plans delivered about 5.4 to 5.6 times the API-equivalent value of comparable ChatGPT plans, while recent price and allowance changes point to a wider question: how long can providers sustain heavily used subscriptions?
The analysis measured how each provider’s usage bar moved across token types, then priced the estimated usage at each provider’s API list rates. Its central comparison covered Claude Opus 5.5 and GPT-6.1 Sol, using a workload dominated by cached input. SemiAnalysis defines API-equivalent value as the plan’s full monthly allowance priced at those list rates; it is an estimate of usage value, not a measure of what subscribers actually consume.
In the report’s figures, a $20 Claude Pro plan represented about $1,178 in API-equivalent usage, against $211 for ChatGPT Plus. At $100, Claude Max 5x was estimated at $5,725 and ChatGPT Pro 100 at $1,055. At $200, Claude Max 20x was estimated at $11,726, compared with $2,084 for ChatGPT Pro 200. SemiAnalysis says the gap remains large when measured in raw tokens, though the models’ different token prices affect the dollar comparison.
The result depends on the selected models and workload. For frontier models, the report found limits were more comparable: it estimated that a $200 OpenAI plan allowance would be exhausted after roughly $2,897 of GPT-6 Astra usage at API prices, while Claude Fable 5.1 would use about half of its plan’s limit at roughly $2,485. The remaining Claude allowance could be used on other models, including Opus or Sonnet.
SemiAnalysis also recorded a recent reduction to OpenAI’s $200 plan: its measured token allowances per model tier were roughly halved. Existing subscribers retain their previous limits until Oct. 29; new purchases receive the lower limits immediately, according to the report. OpenAI added a $500 tier, which the analysis says offers about 21% more Astra than the former $200 plan, but less Sol-class API-equivalent value following a price cut for cached GPT-6.1 Sol input. The new tier’s advertised draw is an Ultrafast mode listed at 300 tokens per second, which SemiAnalysis says it is still testing.
Anthropic’s changes also paired lower API prices with varying allowance increases. SemiAnalysis says Fable 5.1’s cache-read price fell 75% from Fable 5, with no accompanying increase in token limits. Opus 5.5’s input and output prices fell 20%, while cache reads fell 60%; reported allowances rose about 20% on Max and 50% on Pro. OpenAI did not raise Sol limits when GPT-6.1 launched, and the report estimates API-equivalent value on the $200 plan fell about 30%.
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 price comparison matters because high usage can make subscriptions costly to serve. SemiAnalysis estimates subscriptions account for about 10% of Anthropic revenue while consuming more than 40% of its inference compute. It says this lowers blended revenue per megawatt by roughly $36 million. The report describes subscriptions as a larger share of OpenAI revenue, though it does not provide a corresponding figure in the supplied material.
SemiAnalysis models the margin pressure under different usage assumptions. Assuming a subscriber uses the entire allowance and API gross margins are 92%, its estimate puts Opus 5.5 subscription gross margin near minus 369% and Fable 5.1 near 1%. At 20% average utilization, those estimates rise to about 6% and 80%, respectively. These are modeled outcomes, not reported company results; actual costs depend on subscriber behavior and workload.
That distinction helps explain why a large advertised allowance may not translate into a permanent benefit. Providers can change limits, prices, and model availability, while customers’ actual usage varies. A lower API price alone does not give subscribers more usage if the plan allowance stays fixed.
As an affiliate, we earn on qualifying purchases.
How Plan Limits Have Shifted
The report compares subscription plans from OpenAI and Anthropic alongside offerings from Meta, SpaceXAI, Cursor, Cognition, Z.ai, MiniMax and Moonshot. Its central figures focus on OpenAI and Anthropic, and on one coding-agent workload with about 96.6% cached input, 2.6% cache writes, 0.4% fresh input and 0.3% output. The results should be read within that workload and the plan limits SemiAnalysis measured.
SemiAnalysis says OpenAI’s previous tier structure gave higher plans progressively more Astra usage per dollar. After the reported reductions, Pro 100, 200 and 500 returned identical tokens per dollar in its comparison. OpenAI also removed “5x more usage” and “20x more usage” multipliers from its pricing page, according to the report. One practical difference remains: the Pro plans do not have a five-hour usage window, which may help people with concentrated bursts of work.
On Anthropic’s side, the report says the value advantage is concentrated in use of Opus and Sonnet, while Fable-only usage could be less costly to serve. SemiAnalysis’s account of the strategy is an interpretation of its pricing and allowance data, not a stated company rationale.
As an affiliate, we earn on qualifying purchases.
Limits, Usage and Margins
The estimates do not establish how much of each subscription allowance customers use on average. SemiAnalysis’s margin calculations explicitly distinguish full utilization from a 20% average-utilization scenario, and real-world costs can vary with model choice and workload. The supplied source material does not include company responses to the analysis or detailed methodology for every provider’s measured limit.
It is also unclear how long the current allowances will remain in place, whether providers will make further changes, or how the new $500 OpenAI tier performs in practice. SemiAnalysis says it is still testing the advertised Ultrafast mode. The report’s comparisons use API list prices and measured usage limits, so they do not by themselves establish each provider’s subscription revenue, realized margin or customer experience across all tasks.
As an affiliate, we earn on qualifying purchases.
Watch for Further Plan Changes
The next concrete date cited in the report is Oct. 29, when existing subscribers to OpenAI’s $200 plan are due to lose their grandfathered limits. New purchasers already receive the reduced allowance, according to SemiAnalysis. The analysis also says testing of the $500 tier’s Ultrafast mode is ongoing.
Subscribers and developers will need updated measurements to see whether providers change limits again as model prices and compute costs shift. Further comparisons will also need actual usage data to show how advertised allowances translate into typical customer consumption and subscription economics.
As an affiliate, we earn on qualifying purchases.
Key Questions
What did SemiAnalysis compare?
It measured token usage limits across AI subscriptions and priced estimated plan allowances at API list rates. Its headline comparison focused on Claude Opus 5.5 and GPT-6.1 Sol on a coding-agent workload.
How large was the estimated Claude advantage?
For the compared mid-tier plans, SemiAnalysis estimated Claude delivered about 5.4 to 5.6 times the API-equivalent value of ChatGPT. The estimate depends on the models, workload, measured limits and API prices used.
What changed in OpenAI’s $200 plan?
SemiAnalysis says the plan’s token allowances per model tier were roughly halved. Existing subscribers keep their old limits until Oct. 29, while new purchases receive the lower limits immediately.
Does the report show what typical subscribers use?
No. It estimates the value of the full monthly allowance and models margins at full usage and at 20% average utilization. It does not establish average subscriber usage.
Why can an API price cut reduce subscription value?
The report prices plan allowances using API list prices. If the API price per token falls while a plan’s token limit stays the same, the allowance is worth less at those rates; subscribers receive more usage only if the limit also rises.
Source: ThorstenMeyerAI.com
