Anthropic Releases Claude Opus 4.8: Dynamic Workflows + Effort Control

Anthropic has upgraded Claude Opus to version 4.8, which surpasses Opus 4.7 in coding and agent benchmarks at the same price. Two important new capabilities are simultaneously launched: Dynamic Workflows - Claude Code can automatically generate and manage hundreds of sub-Agents in parallel, upgrading complex engineering tasks from "single-threaded" to "multi-threaded parallel"; Effort Control - claude.ai users can independently control the depth of Claude's thinking. The fast mode runs at 2.5 times faster and the price is only 1/3 of the previous generation.

Dynamic Workflows

Claude Code's new Dynamic Workflows feature allows automatic generation and parallel management of hundreds of sub-agents when dealing with very large-scale problems - upgrading from "single-threaded" to "multi-threaded parallel", significantly accelerating the delivery of complex projects.

Effort Control

claude.ai users can independently control Claude's "effort level": choose fast mode for simple queries, and choose deep mode for complex analysis. Fast mode runs 2.5x faster for 1/3 the price.

Benchmark performance

Opus 4.8 surpasses the previous generation in SWE-Bench Pro (64.3%), Terminal-Bench 2.0 and other benchmarks, and early testers reported that its judgment in Agent tasks is more reliable.

Dynamic Workflows and Effort Control represent the evolution direction of AI tools from "single-threaded dialogue" to "distributed parallelism + user controllable". The parallel processing of hundreds of sub-agents and the ability of users to independently adjust the depth of thinking are redefining the collaboration model between developers and AI. For competing products such as GPT-5.5 and DeepSeek, the pace of follow-up in these two directions deserves close attention.

It is worth tracking in the future: the scalability performance of Dynamic Workflows in enterprise-level scenarios, the user adoption rate and actual usage pattern of Effort Control, and the evolution of the competitive situation with GPT-5.5 in terms of Agent capabilities.