Anthropic released Claude Opus 4.8 on May 28, its most capable generally available model to date. The headline feature for agent builders: Dynamic Workflows, a research preview that enables Claude Code to orchestrate hundreds of parallel subagents for codebase-scale migrations from kickoff to merge.
What Shipped
Opus 4.8 arrives just 41 days after Opus 4.7, according to TechCrunch. That accelerated cadence follows a lukewarm reception to 4.7 and mounting competitive pressure from OpenAI’s Codex and Google’s Gemini 3.5 Flash.
The benchmark improvements skew toward agentic workloads. Agentic coding scores rose from 64.3% to 69.2%, knowledge work from 1753 to 1890, and multidisciplinary reasoning with tools from 54.7% to 57.9%, per Anthropic’s documentation.
Pricing remains unchanged from Opus 4.7.
Dynamic Workflows: The Agent Orchestration Layer
Dynamic Workflows is the most significant addition for teams running autonomous agents. The system allows larger models to decompose complex tasks and distribute them across hundreds of parallel subagents, each handling a portion of a large-scale migration or refactoring operation.
“Claude Code alongside Opus 4.8 can now carry out codebase-scale migrations across hundreds of thousands of lines of code from kickoff to merge, with the existing test suite as its bar,” Anthropic wrote in its launch announcement.
The feature is in research preview. Production availability has not been announced.
Cost Reduction for Agentic Loops
Two changes directly reduce costs for long-running agent sessions:
Mid-conversation system messages. Developers can now inject updated instructions mid-task without restating the full system prompt. This preserves prompt cache hits on earlier turns and cuts input token costs during agentic loops. No beta header required.
Lower prompt cache minimum. The minimum cacheable prompt length drops to 1,024 tokens, down from Opus 4.7’s threshold. Shorter agent prompts that previously could not create cache entries now qualify automatically.
Fast Mode: 2.5x Speed, 3x Cheaper
Opus 4.8’s fast mode outputs tokens at 2.5x the standard rate and costs one-third what it did on prior models, according to 9to5Mac. For agent workflows where latency compounds across dozens of tool calls per session, that speed improvement translates to meaningfully shorter end-to-end execution times.
Effort Controls and Honesty Improvements
The default effort level is set to “high” across all surfaces, including Claude Code. Teams can push to “extra” or “max” for difficult tasks and long-running asynchronous workflows, with Anthropic increasing rate limits to accommodate the higher token consumption.
Bridgewater Associates, cited as an early tester, noted that “the biggest difference in the upgrade was Opus 4.8’s tendency to proactively flag issues with the inputs and outputs of an analysis, something other models routinely missed and left to the users to catch,” per TechCrunch.
The Mythos Signal
Anthropic hinted that its more advanced Mythos-class model, held back since an April preview over cybersecurity concerns, may reach general availability “in the coming weeks” once safeguards are complete.
The Orchestration Race
Dynamic Workflows positions Anthropic directly against OpenAI’s Codex multi-agent capabilities and Google’s Gemini agent orchestration layer. The competitive question is no longer which model scores highest on isolated benchmarks, but which platform can reliably coordinate dozens of agents against a shared codebase without requiring external orchestration frameworks. Opus 4.8 is Anthropic’s answer: bring the orchestration inside the model provider’s stack.