Anthropic has launched the Claude Opus 4.8 model, described as a “more effective collaborative partner” with significantly developed capabilities for “Agentic Tasks,” allowing it to plan, analyze complex interdisciplinary problems, and control computer interfaces. Early testers noted the model has become “more honest,” tending to admit uncertainty instead of making up baseless claims. The model is 4 times less likely to pass along code containing errors without flagging them compared to previous versions. On the SWE-Bench Pro programming test, Claude Opus 4.8 achieved a score of 69.2%, outperforming OpenAI’s GPT-5.5 and Google’s Gemini 3.1 Pro, though GPT-5.5 still holds the edge in terminal-coding tests. The “Fast mode” in Opus 4.8 operates 2.5 times faster than previous versions and has become three times cheaper, while regular usage prices remain unchanged compared to the previous Opus 4.7 version. Anthropic also introduced new features including Dynamic Workflows, available as a beta for researchers and subscribers to Business, Team, and Max plans.