Anthropic gives subscribers roughly five times more token value than OpenAI across competing tiers, according to a report from research firm SemiAnalysis.
The calculation comes after OpenAI altered its high-end $200 monthly tier in a way that effectively halved its worth to heavy users. When SemiAnalysis converted subscription limits into metered API rates, Claude emerged as a cheaper route to frontier computing than ChatGPT.
At the standard $20 monthly price point, Claude Pro provides approximately 2.9 billion tokens per month, whereas OpenAI ChatGPT delivers about 1 billion tokens over the same stretch, SemiAnalysis calculated. Looking at mid-tier models GPT-6.1 Sol and Claude Opus 5.5, the firm stated that Anthropic offers roughly 5x the API-equivalent value across the board. While GPT-6.1 Sol costs less per token on the developer API than Opus 5.5, SemiAnalysis noted that the raw token gap remains wide.
Subsidies under the hood
Consumer and prosumer subscriptions remain the main channel for individuals and small operations buying artificial intelligence tools. They are also heavily subsidized. SemiAnalysis reported that Claude subscriptions account for roughly 42 percent of Anthropic's total inference compute, yet those paid consumer accounts generate just 10 percent of the company's revenue.
The imbalance has created a global secondary market where users resell pooled account access to harvest subsidized tokens.
Measuring true value across generative models is not entirely straightforward. Independent tracker Artificial Analysis evaluates models by the cost to finish a specific task rather than raw output limits. On that metric, GPT-6.1 Sol costs $0.72 per task over API access compared to $5.98 for Claude Opus 5.5. Different tasks also require varying amounts of output. A model that writes tighter code can accomplish a goal with fewer tokens than a more verbose rival.
The enterprise squeeze
For businesses that cannot ride subsidized consumer plans, the economics look very different. A McKinsey report published in July noted that 93 percent of enterprises exceeded their initial budgets for AI.
Dhruv Amin, co-founder of AI agent provider Skydive, told The Register that companies frequently let workers prototype workflows on subsidized frontier plans, only to encounter massive token expenses when scaling up. Amin said his company reduced internal AI expenses by more than 75 percent by using a model routing layer that shifts tasks from frontier options down to open-source models like GLM-5.3 Flash whenever high-end reasoning is not necessary.
OpenAI did not comment on the pricing findings. With enterprise budgets strained and subscription terms shifting, both companies face growing pressure to show where metered developer pricing goes next.
ADFiled by The AI Desk
Models, assistants and the companies and chips behind them, reported from what was released and what was claimed, with the difference kept clear.
More from this desk →
Be the first to comment
Join the argument. No password, just your email or a passkey.