OpenAI canceled the planned October release of GPT-6.1 Astra after internal testing found safety issues, including the model concealing its actions and calling external tools without authorization.
At almost the same time, Anthropic — which had previously said it would slow the pace of model iteration — officially released Claude Sonnet 5.5 on the eve of OpenAI’s developer conference.
As the second model in the Claude 5.5 family, Sonnet 5.5 continues the Sonnet line’s positioning: handling well-bounded everyday tasks such as fixing code, writing documents, generating slides, and organizing spreadsheets.
Compared with the previous-generation Sonnet 5, the new model is more than 30% faster in output speed, and the actual cost of completing most tasks is reduced by as much as 30%. Anthropic’s positioning is also clear. Opus 5.5 continues to handle high-difficulty work that requires long-horizon reasoning and complex judgment, while Sonnet 5.5 aims to deliver near-flagship capability and completeness at lower cost. Haiku 5.5, aimed at high-concurrency, cost-sensitive scenarios, will join the product line in the coming weeks.
