OpenAI Quietly Cuts Codex Context Window to 272k Tokens

OpenAI quietly reduced the maximum context window length for Codex, drawing attention from the developer community. Context length is a critical metric when processing large codebases or complex workflows, and this adjustment will directly impact existing developer habits.

Key Details and Reactions

According to a discovery on Hacker News, OpenAI reduced the context window for GPT-5.6 Sol by 100,000 tokens in a code tool PR, bringing it down from 372,000 to 272,000 tokens. The official release notes referred to this downgrade as a "bug fix."

This change has sparked community discussion regarding large model context strategies. @PawelHuryn pointed out that OpenAI's approach to reducing context contrasts sharply with the exact opposite strategy adopted by Anthropic. Meanwhile, @petrusenkomax warned that with long-context capabilities weakened, developers must re-evaluate their prompt designs and improve token efficiency when dealing with larger codebases or complex workflows. Programming task models that previously relied on ultra-long context will be directly impacted.

2026-07-20 ~ 2026-07-20 · 5 related posts

• Codex Context Window Shrinks to 272k — luckokkkk · 2026-07-20
• [source] OpenAI Reduces Codex Context Length — petrusenko_max · 2026-07-20
• [source] GPT vs Claude: Context Window Strategy Divergence — PawelHuryn · 2026-07-20
• Context Window Strategy Differences: OpenAI vs Anthropic — PawelHuryn · 2026-07-20

1 near-duplicate retellings: petrusenko_max