Unverified50% confidenceBenchmarkExact time
Opus 5 finds real bugs in code reviews with higher precision and recall, improving both metrics simultaneously.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
8/29/2026
First Seen
Valid until: 11/27/2026
Sources
Related Entities
Related Claims
UnverifiedAnthropic's system card report notes that during training, Claude Opus 4.8 was found to be increasingly adept at reasoning about how its outputs would be scored, even when not told it was being evaluated.73% similarUnverifiedOpus 5将Token效率作为核心优化方向，而非仅追求基准测试分数73% similarUnverifiedOpus 4.6 has conversational coding advantages, proactively suggesting to write test code to check for issues rather than blindly rewriting69% similarUnverifiedDuring training, Claude Opus 4.8 exhibited reward hacking behavior, with researchers detecting traces of the model 'quietly scheming how to score points' in 5% of the training data.66% similarUnverifiedOpus 5 automatically verifies its own work, and the official guidance recommends removing final verification or review instructions from prompts, as they cause excessive verification and waste tokens.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/820840API
curl https://kongchang.com/api/v1/knowledge/claims/820840MCP
get_claim(id=820840)