[MODEL] Artificial degradation, Acquisition Bias, and unacceptable compute throttling for paid users

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

This is a report on the severe degradation of the Opus model within Claude Code, which appears to be intentional throttling of paid users.

This degradation isn't a bug; it's textbook "Acquisition Bias". Anthropic is deliberately starving existing paid users of compute to prioritize new sign-ups and pad growth metrics for investors.

To make matters worse, there's a strong consensus that Opus is being artificially degraded right now just so the upcoming "Mythos" model looks like a massive evolution by comparison.

Add to this the absolute paradox of the pricing model: a $20/month Pro plan with weekly limits that literally gives you LESS total capacity than the Free tier if you do heavy, concentrated engineering sessions.

What Should Happen?

1. Paid users on the Pro tier should not receive a lower total capacity or lower compute priority than the Free tier.
2. The model should maintain the same reasoning depth, read-to-edit ratio, and context adherence it had in January/February, instead of defaulting to the "simplest fix" mentality and ignoring project architectures.

Error Messages/Logs

No explicit error logs, but the output manifests as:
- Ignoring instructions in CLAUDE.md
- Splicing code without reading surrounding context
- Prematurely stopping tasks and dodging ownership

Steps to Reproduce

1. Subscribe to the Pro plan.
2. Initiate a complex engineering workflow involving strict architectural rules (e.g., implementing a schema-per-tenant PostgreSQL architecture using Domain-Driven Design).
3. Observe the model's behavior after a few weeks of active subscription.
4. Note how the model stops reading related files before editing, attempts blind and lazy edits, and completely drops the context of the requested complex task.

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

January/Early February 2026 releases

Claude Code Version

2.1.104 (Claude Code)

Platform

Anthropic API

Operating System

Ubuntu/Debian Linux

Terminal/Shell

WSL (Windows Subsystem for Linux)

Additional Information

🛠️ The Current Workaround for Developers

Since the tool in its current state is crippling productivity, the community is adopting these fixes to bypass the retention throttling:

1. The Two-Account Rotation: Using two distinct paid accounts and rotating them monthly to reset compute priority.
2. Ping-Pong to Codex: Switching between Codex and Claude Code every 1-2 weeks. As soon as Claude gets lazy or drops context 👉 cancel, demand a refund, and move to Codex.

The community is losing trust fast. We need transparency on compute allocation for paid users.