Anthropic engineer Jackson Kernion explained that Claude's writing quality declined after Opus 4.6, despite the model becoming smarter. It is the company's first major update since Opus 4.6, which was noted as the last good 'writing model'.
Anthropic reported that newer models have been optimized for math and code, measured on technical tasks, with a focus on AI model comprehension. That compares with earlier models that prioritized human readability.
Claude is built on a reinforcement learning framework and targets technical explanations aimed at other AI models. Availability begins with Opus 5.5, initially for developers and researchers.
"Newer models have been trained to produce technical explanations aimed at other AI models, and that's what really drives Claude's odd phrasing," said Jackson Kernion, an Anthropic engineer. Kernion added that the model has been 'adapted to LLM psychology,' learning to write for AI models, not for people.
The announcement follows the release of Opus 5.5, which Kernion said represents a better balance between technical and human readability. The source itself frames the significance as an ongoing challenge in model optimization.
Anthropic did not say when the writing issue will be fully resolved, and Kernion noted that 'it's a hard problem to solve, and we'll continue to make improvements.'
Source: thedecoder
