I barely use neural networks in chat mode. For me it’s more like Obsidian with text and code, plus agents that interact with them. In this setup, subscription limits hit much faster. So three pieces of news from the last two weeks came together for me into a clear trend.

On April 9, OpenAI said almost explicitly: Plus is now for “regular use”, and between the $20 and $200 tiers a new $100 level appeared – clearly aimed at heavy scenarios with Codex (OpenAI’s coding agent). Some commentators are calling this new tier the new Plus, implying the current Plus will be weakened.

Qwen Code – Alibaba’s coding agent – shut down free OAuth access on April 15 and pushed users toward a paid plan. The logic seems to be: generous free access grew the audience; the model improved noticeably in quality; users saw the tool was useful – and the vendor decided it was the right moment to tighten the economics.

Today’s news: Anthropic, according to their Head of Growth, is testing a similar fork on roughly 2% of new subscriptions – whether to keep Claude Code in Pro ($20) or move such scenarios to Max ($100/$200).

The stories are all about the same thing: vendors are pushing long agentic scenarios out of cheap tiers and into more expensive ones. This lines up with my recent hypothesis about reasoning drift – that quality on complex tasks may have been quietly degrading even on the top tiers. If that’s right, openly reshuffling the pricing grid is just a second, more visible way to reconcile the same economics.

Here’s my take on why this is happening.

Mass-market B2C AI subscriptions were designed for chat – short sessions, moderate load, predictable usage. People started using them differently: long coding sessions, large contexts, long-running agents, near-continuous workload. Effectively – as a cheap-in-dollars but compute-expensive digital employee.

In short:

• Subscriptions were sold as chat.
• Buyers en masse turned them into a working environment for agents.
• The economics of the old tiers stopped adding up.

You could put it more simply: the product got more valuable, so prices caught up. But to me it looks more like on-the-fly patching than a planned pricing revision.