Claude Opus 4.8: Misalignment Rates & Performance Boost

Claude Opus 4.8 shows misalignment rates similar to Claude Mythos Preview. This new model from Anthropic replaces Opus 4.7 starting today, offering faster thinking at one-third the cost. Opus 4.8 scores higher than 4.7 on two coding benchmarks. Anthropic states that Opus 4.8 has substantially lower rates of misalignment than its predecessor. The company emphasizes model safety and interpretability with this release, claiming Opus 4.7 had a 92% honesty rate. OpenAI also released GPT-5.5 Instant, which is less verbose and has fewer hallucinations than GPT-5.3 Instant. It produced 52.5% fewer hallucinated claims on high-stakes prompts. The bottom line is that AI labs are continually shipping new models, each with specific strengths and improvements in areas like safety and performance.