Anthropic's newly released model Opus 4.8 has been put to the test in early-access testing. According to reports, Opus 4.8 excels in creating greenfield prototypes and one-shot features, with fast execution times. However, it struggles with the "last 10%" problem, where tasks require a high level of precision and attention to detail. Additionally, the model had difficulty handling edge cases in existing codebases and was prone to hallucinations.
In comparison to its predecessor Opus 4.7, Opus 4.8 showed improvements in business strategy work, but still lags behind in data-heavy tasks such as roadmapping. New features have been added to Claude.ai and Cowork, including dynamic workflows with parallel subagents and effort control.
Opus 4.8 is available for testing, with pricing information not yet disclosed. The model has sparked interest among entrepreneurs and founders, who are eager to explore its capabilities and limitations.
