Claude Code just got WAY more expensive (called it)

Introduction & Background 00:00

  • Discusses recent reactions to Cursor's price changes and the shift of users to Claude Code for perceived better value
  • Warned previously that low pricing in AI tools wouldn't be sustainable due to high GPU and inference costs
  • Announces that Anthropic is implementing weekly usage limits for Cloud Pro and Max users starting end of August

The Business Economics of Claude Code Pricing 00:28

  • Super power users have been running Claude Code continuously, costing Anthropic tens of thousands of dollars in some cases
  • Example cited of a single user consuming tens of thousands in model usage on a $200/month plan
  • Unlimited access for a fixed payment is unsustainable; some users would pay much more if allowed but still wouldn't cover provider costs

Code Rabbit Sponsorship & Role of AI in Code Review 01:16

  • Endorses Code Rabbit, an AI-driven code review tool, describing its usefulness, efficiency, and integration with VS Code
  • Notes benefits like fast, detailed code reviews, automatic comments on bugs, and improved review workflow
  • AI helps streamline syntax and bug catching, letting human reviewers focus on understanding code intent

Claude Code's Development and Usage Trends 03:37

  • Reviews history: Claude 3.5 Sonnet's release led to improved code tools and explosion of agentic coding solutions
  • Original Claude Code creators left for Cursor, after which Claude Code's popularity sharply increased, particularly among heavy users
  • Introduction of a $200/month Max tier shifted users from API-based pricing to flat subscriptions

AI Model Cost Structures & Comparison 05:00

  • Explains model pricing based on input and output tokens, with output generally more costly
  • Claude models (e.g., Claude 4) are significantly more expensive than alternatives like Gemini or Groq, both per token and in total usage costs
  • Some models (like Groq 4) generate an excessive number of output tokens, inflating costs further
  • Usage inefficiency can lead to single commands costing $12–15, often without satisfactory results

Subscription Model Flaws & User Abuse Patterns 08:21

  • The fixed-price $200 tier led to device clusters and intensive, often 24/7, usage by “power users”
  • Anecdotes of users running Claude code for extensive projects and experiments, consuming massive token counts daily
  • High-end users constitute under 5% but, given a large user base, this represents thousands exploiting the system

Zero Interest Rate Phenomenon & AI Tool Subsidies 12:23

  • Compares the period of cheap Claude Code access to the startup-friendly "ZERP" era, where companies spent heavily to acquire users before worrying about costs
  • Suggests Anthropic's approach may have been a calculated bet to buy market share, akin to large-scale marketing spending

Anthropic’s Announcement & Industry Response 15:03

  • Anthropic will limit usage for Pro and Max tiers, estimating impact on fewer than 5% of users
  • Abusive scenarios include excessive model usage and sharing/reselling accounts
  • Option for Max users to purchase extra usage at standard API rates
  • Users complain that API rates are too expensive compared to previous subsidized offerings

Real Cost of Large-Scale AI Inference 16:11

  • Explains high costs for running advanced AI models due to GPU, energy, and infrastructure needs
  • Providers cannot sustainably offer AI at rates far below these costs without losing money or sacrificing capacity for higher-paying clients

Cursor & Claude Code Pricing Missteps 17:02

  • Outrage over price hikes not due to greed but due to fundamental business sustainability issues
  • Anthropic underestimated the costs users would incur under the $200/month model, now resulting in a PR hit

Concrete Examples of Expensive AI Code Generation 17:27

  • Reports cases where users generated tens of thousands of lines of code or massive files with single commands, costing hundreds or thousands of dollars under real pricing

Emerging Cheaper Alternatives 17:59

  • Mentions Open Code and other models (e.g., Qwen 3 Coder) offer comparable or better performance for less cost

Final Takeaways & Lessons 18:24

  • “If it’s too good to be true, it’s subsidized”—cheap unlimited AI comes at substantial provider expense and is not sustainable
  • Cheaper-than-API pricing is usually marketing to gain users, not a long-term business model
  • Unlimited offers in AI are impossible due to real physical and financial constraints on compute/electricity/GPU availability
  • Encourages feedback and discussion on adaptation to the new Claude Code pricing