Users Report Degradation in OpenAI Sol High Chat/Codex
Illustrious-Bet-1368 · reddit · 2026-08-22
Reddit users report recent degradation and changes in OpenAI's Sol High models (Chat and Codex):
Chat (Sol High):
- Reasoning chains have disappeared; replies are instant but lack depth.
- Quality dropped with swearing/emojis; sources (memories/instructions) are no longer shown.
- Title generation bugs (e.g., outputting "Write X Reply").
Codex:
- Stricter usage limits.
- Fails to follow instructions/docs, doesn't pull Memory or Skills, suggests generic implementations.
- Ignores explicit testing constraints, over-engineering.
Users suspect this is a widespread A/B test or model adjustment.
More from Models
- Grok Bot Ran a 12-Hour Data Analysis Nonstop, Handling API Rate Limits Itself — Daniel_Farinax · 2026-10-16
- One year of Recursive Language Models: GPT-5-mini beats GPT-5 by 110% on 132k-token tasks, cheaper too — a1zhang · 2026-10-16
- Users Report Claude Models Suddenly More Compliant After Forced Compaction — RealSharpNinja · 2026-10-16
- Prediction: Opus 5.5-level models on laptops by 2027, backed by Qwen 27B data — cephaloform · 2026-10-16
- Users Say Krea 2 Output Degraded Sharply Since Krea 3 Beta Launched — szywis · 2026-10-16
- LLM-as-a-Verifier lifts Nemotron 3.5 Super by 9-10 points on agentic benchmarks — simonguozirui · 2026-10-16