AI pricing shifts to power users
OpenAI cut its ChatGPT Pro price to $100/month targeting heavy Codex users, and industry reporting shows both OpenAI and Anthropic are segmenting plans with lower entry tiers and richer mid‑tiers for intense users. Microsoft executives have even suggested AI agents could require licensed 'seats' like human employees — a shift that would recast AI economics toward per‑agent or per‑workflow billing rather than per‑employee. (lifehacker.com) (siliconangle.com) (businessinsider.com)
OpenAI just filled in the weirdest gap in consumer artificial intelligence pricing: until this week, ChatGPT jumped from Plus at $20 a month to Pro at $200, and now there is a new Pro tier at $100 aimed at people who use Codex for long coding sessions. (openai.com) That new $100 plan gives 5 times the limits of Plus, while the $200 plan stays in place with 20 times the limits of Plus, which means OpenAI is no longer charging one “professional” price but three distinct prices for three different levels of intensity. (openai.com) Codex is the part of ChatGPT built for software work, and OpenAI has been tuning its pricing around that workload instead of around general chat, including a rate-card change on April 2, 2026 that moved Codex billing toward token usage for several plans. (openai.com) OpenAI also added pay-as-you-go pricing for Codex teams last week inside ChatGPT Business and Enterprise, which gives companies a way to start with no fixed monthly Codex commitment and then scale usage as engineers actually use it. (openai.com) Anthropic is moving the same direction from the other side of the market: its Claude lineup now spans Pro, Max, Team, and Enterprise, and its business plans include premium seats, extra-usage options, spend caps, and usage analytics instead of one flat “all power users pay this” offer. (anthropic.com) Anthropic’s latest model pages also show how that ladder works in practice, with Claude Opus 4.6 available across Pro, Max, Team, and Enterprise rather than being locked behind a single top plan. (anthropic.com) The pattern is simple: casual users get a cheaper on-ramp, while heavy users get mid-tier plans with bigger limits, and companies get admin controls that turn usage into something finance teams can meter, cap, and approve. (siliconangle.com) Microsoft is pushing the next step past human subscriptions and into agent subscriptions, because its April 2026 Copilot Studio licensing guide is already framed around “agents” as billable units with their own licensing rules and payment methods. (microsoft.com) Business Insider reported this week that a Microsoft executive went further and suggested artificial intelligence agents may need software licenses the way employees need seats today, which would turn an agent from a feature inside a worker’s account into a customer of software in its own right. (businessinsider.com) If that model sticks, a law firm would not just pay for 500 lawyers to use software; it could pay for 500 lawyers, 200 research agents, 50 drafting agents, and a separate meter for the workflows those agents run all day. (businessinsider.com) That is why the $100 ChatGPT tier is bigger than one price cut: OpenAI, Anthropic, and Microsoft are all testing a market where the basic question is no longer “how many employees use artificial intelligence,” but “how much autonomous work gets done, by whom, and on whose meter.” (openai.com)