Claude-Code Vs. Codex — Part 1

· Source: Towards AI - Medium · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, quick

Summary

Anthropic's Claude-code, a core product, has recently experienced a degradation in code quality and user experience, contrasting with the improving desktop application for OpenAI's Codex. Over a six-week period, three distinct changes impacted Claude's performance: a silent reduction in reasoning effort, a caching bug causing forgetfulness in extended sessions, and a verbosity cap introduced concurrently with the Opus 4.7 release, complicating performance analysis. Additionally, a server-side filter misfired, rerouting Max plan requests to pay-as-you-go billing, and the system prompt is undergoing quiet modifications. These issues have led to a shift in user sentiment, with the author noting a potential change in preference towards Codex within three months.

Key takeaway

For Machine Learning Engineers evaluating code generation tools, be aware that Claude-code's recent performance issues, including reduced reasoning and billing errors, suggest a need for caution. Your current preference for Claude may shift as OpenAI's Codex desktop app continues to mature. Consider re-evaluating your primary AI coding assistant in the next quarter to ensure optimal productivity and cost efficiency.

Key insights

Recent changes to Claude-code have degraded its performance and user experience, contrasting with Codex's improvements.

Principles

In practice

Topics

Best for: CTO, Machine Learning Engineer, VP of Engineering/Data, AI Engineer, Software Engineer, Director of AI/ML

Related on AIssential

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Towards AI - Medium.