Anthropic Subscriptions Offer 5x+ More Value Than OpenAI
4 hours ago
- Subscription plans are heavily subsidized, serving as customer acquisition tools that impact margins despite representing a small portion of revenue.
- The value of a plan depends on the (plan, model, workload) tuple, requiring dynamic tracking due to frequent limit changes by labs.
- Methodology measures token rates by isolating token types via experiments, tracking meter movements, and computing API-equivalent value.
- Anthropic currently offers significantly better value for mid-tier models (e.g., Opus 5.5) than OpenAI (e.g., GPT 6.1 Sol), with ~5x API-equivalent value.
- OpenAI recently halved the $200 plan value and introduced a $500 plan, equalizing tokens per dollar across Pro tiers.
- Labs adjust subscription limits inconsistently after API price cuts; Anthropic increased Opus limits but not enough to offset cuts.
- Two strategies for reducing subsidies: Anthropic subtly reduces value for premium models; OpenAI abruptly cuts limits across all models.
- Chinese labs still offer subsidized plans with varying value per dollar; third-party wrappers provide worse value than first-party plans.