I compared Claude Opus 4.8 with 4.7 in a 10-round honesty test - and a legal prompt broke it
The comparison between Claude Opus 4.8 and 4.7 revealed that the newer model performs better in terms of honesty and judgment. However, a significant judgment error in Opus 4.8 indicates that there is still room for improvement. The testing involved various prompts across different domains, with results cross-checked by multiple AI systems.
- ▪Claude Opus 4.8 was tested against its predecessor, Opus 4.7, in a series of prompts designed to expose weaknesses.
- ▪The tests included coding, medical, finance, and legal scenarios to evaluate the models' responses.
- ▪Overall, Opus 4.8 showed improved honesty and calibration compared to Opus 4.7, despite a notable judgment error.
Opening excerpt (first ~120 words) tap to expand
Innovation Home Innovation Artificial Intelligence I compared Claude Opus 4.8 with 4.7 in a 10-round honesty test - and a legal prompt broke it The latest models were pitted against coding, medical, finance, and legal traps, then I cross-checked the results with multiple AIs. Written by David Gewirtz, Senior Contributing EditorSenior Contributing Editor June 2, 2026 at 6:07 p.m. PT David Gewirtz/ZDNETFollow ZDNET: Add us as a preferred source on Google. ZDNET's key takeaways Claude Opus 4.8 handled uncertainty better than 4.7.Multiple AIs helped cross-check the test results.Even honest AIs can still rationalize bad assumptions.Last week, Anthropic released its latest frontier large language model, Claude Opus 4.8.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at ZDNet.
Discussion
0 commentsMore from ZDNet
