GPT-5.5 Codex reasoning-token clustering may be leading to degraded performance

What happened

A broad GitHub issue #30364 was opened in the OpenAI Codex repository documenting community concerns about degraded output quality from GPT-5.5 Codex, gathering 107 Hacker News points on July 4, 2026.

Context and impact

GPT-5.5 Codex was considered a significant coding improvement over GPT-5. If the degradation is confirmed, it may indicate regression caused by RL post-training — a known pitfall with aggressive RLHF fine-tuning. OpenAI has not yet issued a statement.

Details

• Issue #30364 in the OpenAI Codex repository, filed July 4, 2026
• HN score: 107 points, author: maille
• Community hypothesis: reasoning-token clustering during RL training process
• Manifests as lower-quality code output and less predictable results
• OpenAI has not issued a public response