Search Issues

Popular Questions

GPT-5 Feels Worse Than GPT-4o: Why and What to Do About It

Many users report GPT-5 (and GPT-5.5) feeling less capable than GPT-4o for creative writing, coding, and following complex instructions. This is not a bug — GPT-5 was optimized for reasoning and safety over raw creative output. Workarounds: use o3-mini for coding tasks, add explicit formatting instructions, or use the system prompt to request GPT-4o-style behavior. GPT-4o is no longer available in the ChatGPT interface but remains accessible via the API.

ChatGPT Memory Not Working: How to Fix Saved Memories Not Persisting

ChatGPT memory stops working when the feature is toggled off in Settings, when you exceed the 256KB memory storage limit, or after major model updates that reset memory indexing. Fix: go to Settings → Personalization → Memory and confirm it is enabled. If enabled but not working, clear old memories to free space, then explicitly tell ChatGPT to 'remember this' in your next conversation.

Claude Code Prompt Caching Bug: How It Inflates Usage 10–20x and the Fix

Claude Code versions before v2.1.34 contained two independent bugs that broke prompt caching, causing every interaction to be billed at full token cost instead of the cached rate. This inflated usage by 10 to 20 times, draining daily Pro budgets in under 20 minutes. The fix is to update Claude Code to v2.1.34 or later with 'npm update -g @anthropic-ai/claude-code'. Anthropic confirmed the bug in March 2026.

Claude Code Usage Limit Draining Too Fast: Causes and Fixes

Claude Code drains your usage limit fast because each tool call (read file, write file, run command) counts as a separate token-consuming interaction, and a known prompt caching bug in versions before v2.1.34 inflated costs by 10–20x. Fix: update Claude Code to the latest version, switch the default model from Opus 4 to Sonnet 4, and break large agent sessions into smaller tasks.