Anthropic's subscription plans offer about five times more value than OpenAI's mid-tier plans, according to a deep dive by SemiAnalysis. Both companies are trimming subsidies ahead of their mega-AI IPOs.
TL;DR
- Anthropic's subscription plans provide significantly more value than OpenAI's, particularly at the mid-tier level.
- Both companies are reducing subsidies on their subscription plans ahead of their upcoming IPOs.
- Subscription plans are a small revenue share but account for a significant portion of inference compute, impacting revenue per megawatt.
What happened
SemiAnalysis conducted a deep dive into the subscription plans of major AI companies, including Anthropic and OpenAI. The analysis found that Anthropic's plans offer about five times more value than OpenAI's mid-tier plans, with the gap holding even when considering raw tokens instead of dollars.
Anthropic has been closing the 'all you can eat' buffet, cutting heavy third-party agent users off its flat-rate plans and launching new models at higher prices. OpenAI, on the other hand, has been cutting prices and increasing usage limits, but recent changes have reduced the value of its plans.
Both companies are preparing for their mega-AI IPOs, with Anthropic aiming to list as soon as mid-November at a valuation of $1.8 to $2 trillion. OpenAI is staying private this year and raising more private capital in the meantime.
Why it matters
Subscription plans are a small share of revenue but a big share of compute, impacting revenue per megawatt and gross margins. Both companies are trimming subsidies to improve margins ahead of their IPOs.
Anthropic's model-by-model approach to reducing subsidies allows it to keep the 'best value' label with developers, while OpenAI's all-at-once approach may invite a backlash but avoids one due to other news and grandfathered plans.
Owning the models is the real moat in pricing, allowing the labs to subsidize where they want, cut where they need to, and still out-earn the apps that resell their models.
Key facts
- Anthropic's top-tier plan offers about $2,485 worth of Fable 5.1 at API prices before hitting the cap, while OpenAI's top-tier plan offers about $2,897 worth of GPT-6 Astra.
- Anthropic's mid-tier plan offers about five times the API-equivalent value of OpenAI's mid-tier plan across the board.
- OpenAI halved the tokens its $200 Pro plan gets for each model tier and added a new $500 tier with an 'Ultrafast' mode, reducing the value of its plans.
- Anthropic's subscription plans are about 10% of revenue but can take up over 40% of inference compute, lowering blended revenue per megawatt by about $36 million.
- OpenAI's subscription plans are a larger share of total revenue, making them even more important for the company's margins.
- Chinese labs offer subsidized plans with value per dollar a little below OpenAI's roughly 12x.
- Third-party AI coding apps like Cursor and Cognition offer much less generous plans than the labs' own.
Context
The AI industry is shifting from price per token to price per task, as companies budget AI tokens and ask what they get for the spend. Subscriptions are the consumer and small business version of that question.
Both Anthropic and OpenAI have been in a pricing tug of war, with Anthropic closing the buffet and OpenAI cutting prices. The recent analysis by SemiAnalysis turns some of the conventional wisdom on its head.
The shift in subscription plans is part of a broader trend in the AI industry, as companies prepare for IPOs and focus on improving margins. Owning the models is a key advantage, allowing labs to set the terms and out-earn resellers.
