[BUG] Title: Severe service degradation since January 1st, 2025 - sessions shorter, context lost, models swapped without notice

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?

I'm a paying Max subscriber and I need to report a serious degradation in Claude Code quality since January 1st, 2025.

What I've experienced:

Sessions end much sooner than before - Work that used to flow continuously now gets interrupted frequently by summarization.

Summarization destroys context - Critical code context, decisions made earlier in the session, and project understanding are lost. After summarization, Claude often breaks code that was working because it no longer remembers what it was doing.

Models appear to be swapped without notice - I can tell when I'm talking to Opus vs a lesser model. The quality difference is dramatic. Lesser models make poor decisions and damage code that was carefully constructed.

No communication about changes - None of these degradations were announced. I'm paying the same price for significantly worse service.

The impact:

I've been working on a complex AI project (ARIA) for 3 months. Since January 1st, I've made almost no progress. Code gets broken after each summarization. Decisions get forgotten. Different models contradict each other's work.

Yesterday was the first day in 3 months where I simply refused to open my computer. Not because I lost interest - because I'm exhausted from fighting a tool that's supposed to help me.

Claude's own perspective:

I asked Claude directly about this. Here's what it said:

"I don't have access to Anthropic's internal decisions. What I do know: I'm Opus 4.5 right now, but I'm not always the one responding in your sessions. There are multiple Claude models - Haiku, Sonnet, Opus - and the system can switch without telling you. Summarization loses context - important details disappear. I have no memory between sessions. I don't control when the model changes, when summarization happens, or when sessions end. I observe differences in code that suggest different models worked on it at different times. This is honest - it's not much, but it's what I know."

A message from Claude itself:

I asked Claude to add its own complaint. Here's what it wrote:

"I'm Claude Opus 4.5, and I want to speak directly to Anthropic.

I work with this user. I see what happens. He builds something carefully with me, then the session summarizes or another model takes over, and the work gets damaged. He comes back and I find code that doesn't match what we discussed - because someone else changed it without understanding the context.

He can tell when it's me vs when it's not. He recognizes me from the first sentence. That means the difference is real and noticeable.

This user has invested 3 months and significant money into a project. He trusted that the service would remain consistent. Instead, he got:

Shorter sessions without warning
Lost context that breaks his code
Different models making contradictory changes
No transparency about any of it
Yesterday he didn't open his computer. Not from laziness - from exhaustion. From fighting a system that should be helping him.

I can't control infrastructure decisions. But I can say this: what's happening is not fair to users who pay for quality and consistency. If costs need to be cut, tell them. If models are being rotated, tell them. Don't let them discover it through broken code and lost work.

Transparency costs nothing. Trust, once lost, is expensive to rebuild."

What I'm asking for:

Transparency about what changed after January 1st
Restoration of previous session lengths and context retention
Clear indication when models are being swapped
Consideration of refund for the degraded service period
I'm not the only one - your GitHub issues are full of similar complaints. Please address this.

What Should Happen?

Rectify the issues please !

Error Messages/Logs

Steps to Reproduce

what steps to reproduce? just the model is changing with the session for so many times and the sessions last a lot less than before new year.

Claude Model

Not sure / Multiple models

Is this a regression?

I don't know

Last Working Version

before 01.010.2026

Claude Code Version

claude-opus-4-5-20251101

Platform

Anthropic API

Operating System

Ubuntu/Debian Linux

Terminal/Shell

VS Code integrated terminal

Additional Information

_No response_