Anthropic Unveils Claude Opus 4.8: A Technological Leap Ahead of Mythos

Anthropic announced the launch of Claude Opus 4.8, a significant update to its artificial intelligence model, on Thursday, May 28. Aimed at businesses and developers, this model promises notable improvements in coding and autonomy, while maintaining the same pricing as its predecessor.

Enhanced Performance in Coding and Professional Tasks

Claude Opus 4.8 demonstrates improved performance, particularly in coding, with a score of 69.2% on SWE-Bench Pro, surpassing GPT-5.5 and Gemini 3.1 Pro. However, on Terminal-Bench 2.1, GPT-5.5 retains an advantage with 78.2%. The model also excels in professional tasks, achieving a score of 1890 on GDPval-AA, the best in the industry. Opus 4.8 is approximately four times less likely than its predecessor to overlook a defect in the code without reporting it. According to early feedback from adopters, the model is said to be worthy of a human collaborator with “its critical sense, taste, and constructive remarks.”

New Features for Developers

Anthropic is introducing several features with Opus 4.8. Effort control is now available on the web version of Claude and with Cowork. The dynamic workflows, launched in preview on Claude Code, allow for the launch of hundreds of sub-agents in parallel, thereby optimizing work sessions. Additionally, the fast mode, which accelerates Opus 4.8 to 2.5 times the normal speed, is offered at a cost three times lower than on previous models.

Future Prospects with Claude Mythos

Anthropic is pursuing two goals: making Opus more accessible in light of the pricing competition from OpenAI and Google, and developing a new class of models with Claude Mythos. Already tested by a few companies, Claude Mythos is expected to be more widely available in the coming weeks.