Debate Erupts Over OpenAI vs Anthropic Default Chain-of-Thought Retention in APIs

steipete · x · 2026-07-31

A heated discussion has emerged in the developer community regarding how LLM APIs handle chain-of-thought (CoT) across multi-turn conversations. Some developers argue that OpenAI's default behavior of not retaining prior reasoning and applying a compaction strategy is an arbitrary decision that could hurt models not expecting this approach, calling it a counterproductive rule if it goes against recommended usage.

However, others countered by pointing out that Anthropic's Claude Opus 5 API by default automatically retains all prior thinking blocks in the conversation context. The community emphasized that benchmarks should allow different models to benefit from the same mechanisms to ensure fair and professional evaluations.

Original post →

More from coding & agent

coding & agent channel →