On May 29, Beijing time, Anthropic officially launched Claude Opus 4.8, just 41 days after the release of its predecessor, Opus 4.7, setting a record for the shortest iteration interval for the company’s flagship model. According to Anthropic’s official announcement, the new model’s agentic coding score (SWE-Bench Pro) rose from 64.3% to 69.2%, and its multi-disciplinary reasoning score increased from 54.7% to 57.9%. In the Online-Mind2Web browser agent test, it outperformed GPT-5.5 and Gemini 3.1 Pro with a score of 84%. Standard pricing remains unchanged at $5 per million input tokens and $25 per million output tokens. In this update, Anthropic highlighted “honesty” as one of the most prominent improvements—internal evaluations show that the probability of Opus 4.8 overlooking flaws in generated code without prompting is approximately one-fourth that of Opus 4.7.
This update also introduces two new features: first, “Effort Control,” which allows users on claude.ai and Cowork to manually adjust the model’s reasoning depth, ranging from low-energy fast responses to high-intensity deep reasoning; and second, “Dynamic Workflows” (research preview) for Claude Code, which supports the parallel scheduling of hundreds of sub-agents within a single task, capable of completing full-repository migrations on the scale of hundreds of thousands of lines of code. Additionally, Fast Mode has been accelerated to 2.5 times its regular speed, with costs reduced to one-third of previous levels. Anthropic also revealed that the more capable Claude Mythos-level models are expected to be available to all customers “in the coming weeks,” but they currently remain in a limited preview stage as cybersecurity protection systems are still being refined.