Claude Opus 4.8: "a modest but tangible improvement" (simonwillison.net)

Anthropic has released Claude Opus 4.8, which the company describes as "a modest but tangible improvement" over its predecessor. This update emphasizes enhanced honesty and reliability, with Opus 4.8 exhibiting a significantly lower rate of factual hallucination; it is approximately four times less likely to overlook flaws in code it generates compared to earlier versions. This reduction in inaccuracies primarily results from the model's tendency to flag uncertainties rather than confidently asserting incorrect information. Despite these improvements, the model maintains its previous pricing structure and characteristic limits regarding token context and output. One noteworthy feature of Claude Opus 4.8 is its ability to accept mid-conversation system messages, allowing users to modify instructions without restating the entire prompt. This capability enhances flexibility during longer interactions and can optimize input costs in iterative scenarios. Additionally, the minimum cacheable prompt length has been reduced, facilitating more efficient use in practical applications. These updates reflect a commitment to improving user experience while maintaining transparency in model performance, indicating a positive direction for ethical AI development within the AI/ML community.