[MODEL]

Preflight Checklist

• [x] I have searched existing issues for similar behavior reports
• [x] This report does NOT contain sensitive information (API keys, passwords, etc.)

Type of Behavior Issue

Claude made incorrect assumptions about my project

What You Asked Claude to Do

Newer models, like Sonnet 4.6, are underperforming when compared the previous models.
Some coding I must using Github Copilot + Sonnet 4.0 to fix, because Sonnet 4.6 can't understand instructions nor figure out a consistent solution. Also sometimes 4.6 seems as verbose and repeting as the Haiku models.

What Claude Actually Did

This is about Sonnet 4.6

1. Output sometimes are verbose or repeating, just like a Haiku model.
2. forget or misunderstand instructions.
3. forget or misunderstand context.

Expected Behavior

1. Work like previous models, like Sonnet 3.7 or 4.0

Files Affected

- issue related to bad model performance, not to specific files

Permission Mode

Accept Edits was ON (auto-accepting changes)

Can You Reproduce This?

Sometimes (intermittent)

Steps to Reproduce

1. Ask some complex task on Sonnet 4.6 - it fails over and over.
2. Ask same complex task on Github copilot + Sonnet 4.0 - it works.

Claude Model

Sonnet

Relevant Conversation

Impact

High - Significant unwanted changes

Claude Code Version

v2.1.46

Platform

Anthropic API

Additional Context

The Sonnet 4.6 sometimes is verbose, repeating steps, just like a Haiku model.