SemiAnalysis limit-tests AI subscriptions
Claude plans give ~5x the API-equivalent value of OpenAI’s at the mid tier
Partly confirmed
The takeaway
On Oct 5, 2026 SemiAnalysis published limit tests of the subscription plans of Anthropic, OpenAI, Meta, SpaceXAI, MiniMax, Moonshot, Z.ai, Cursor and Cognition.
Status
- Claim
Partly confirmed
- Our reporting
- Medium confidence
- Importance
- 2 of 5
- Last verified
- 7 October 2026
Your AI and this story
- GPT-6.2 Astra127 days after its cutoff
- Claude Opus 5.597 days after its cutoff
- Gemini 4 Hydrogen188 days after its cutoff
- Grok 4.7127 days after its cutoff
None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 97 days before it.
Key facts
- Method: measured how far each plan’s usage meter moves per token type (input, cache write, cache read, output) over 5-hour and weekly windows, then converted the limits into API-price equivalents (SemiAnalysis; authors Andrew Megalaa, Max Kan, Dylan Patel)
- ‘At this mid-tier of models… Anthropic is an overwhelmingly better deal, offering ~5x the API-equivalent value across the board’; OpenAI’s only offset is that its Pro plans have no 5-hour limit, which SemiAnalysis says does not offset ‘the ~4x higher API-equivalent value offered by Opus 5.5’
- ‘Despite being just 10% of overall revenue, subscriptions can take up over 40% of inference compute and lower blended revenue per MW by ~$36M’
- Context: OpenAI recently halved limits on its $200 plan and added a $500 Pro tier; SemiAnalysis credits OpenAI’s generous limit resets with part of Codex’s adoption surge and says Anthropic has repeatedly walked back planned subscription cuts
- The Register (Oct 6): Claude Pro ($20/month) yields ~2.9B tokens a month of API-equivalent usage vs ~1B for ChatGPT at the same price; neither company commented
What happened
SemiAnalysis ran controlled usage tests on consumer and developer subscription plans from nine AI providers and turned each plan’s limits into the dollar value the same tokens would cost on the API. Its headline: at the mid tier, Claude subscriptions deliver about five times the API-equivalent value of ChatGPT/Codex subscriptions. Techmeme picked it up on Oct 5.
Why it matters
Subscription limits, not list API prices, now decide much of the competition for coding-agent users. The finding that subscriptions can take over 40% of inference compute while bringing ~10% of revenue explains why labs keep changing their limits.
Unverified: the second half of the report is paywalled, so per-plan numbers (including the Fable 5.1 vs GPT-6 Astra comparison) were not read. The “$500 tier adds only 21% value” figure seen in a summary was not confirmed in the free text and is omitted.
Sources
3 sources from 3 sites. Numbers match the chips in the text.
3 sources: 2 press, 1 reaction
Press
- The Register: Anthropic Claude subscription plan provides more value than OpenAI’s, study saystheregister.com, press
- SemiAnalysis (Oct 5): Anthropic subscriptions offer 5x more value (partly paywalled)newsletter.semianalysis.com, press
Reactions
- Lisan al Gaib (@scaling01) on X: ‘this is brutal’ (~144k views)x.com, discussion
Changes
- Added The Register’s write-up
- Filed