Anthropic 推出 Claude Opus 4.8，同步發布動態工作流程與更便宜的快速模式，工作流程最多支援 1,000 個子代理

Anthropic just launched Claude Opus 4.8. Also, there two Claude Code updates shipped with it. Dynamic workflows run many subagents in parallel. Fast mode now supports Opus 4.8 at a lower price. Both are research previews. What Dynamic Workflows Actually Are A dynamic workflow is a JavaScript script that orchestrates subagents at scale. Claude writes the script for the task you describe. A runtime then executes it in the background. Your session stays responsive while agents work. The plan moves into code, not Claude’s context window. Intermediate results live in script variables instead. So Claude’s context holds only the final answer. That is the core difference from subagents and skills. The feature requires Claude Code v2.1.154 or later. It runs in the CLI, Desktop, and VS Code extension. It is available on Max, Team, and Enterprise plans. It is on by default on Max and Team. On Enterprise it is off until an admin enables it. It also runs on the Claude API, Amazon Bedrock, Vertex AI, and Microsoft Foundry. You can start a workflow in two ways. Include the word workflow anywhere in your prompt. Or turn on a setting called ultracode. Ultracode combines xhigh reasoning effort with automatic workflow orchestration. Claude Code also bundles /deep-research as a built-in workflow. How a Workflow Runs When a workflow kicks off, Claude plans dynamically from your prompt. It breaks the task into subtasks. It fans work across subagents running in parallel. Agents address the problem from independent angles. Other agents then try to refute those findings. The run iterates until the answers converge. Results are checked before they are folded in. The runtime applies hard limits. It allows up to 16 concurrent agents. It caps each run at 1,000 agents total. The workflow script itself cannot touch the filesystem or shell. Only the agents read, write, and run commands. Progress is saved as the run proceeds. An interrupted job resumes within the same session. Completed agents return cached results on resume. Coordination happens outside the conversation, so the plan stays on track. The Bun Rewrite Example Anthropic highlighted one large example. Jarred Sumner used dynamic workflows to port Bun. The port moved Bun from Zig to Rust. It passed 99.8% of the existing test suite. It produced roughly 750,000 lines of Rust. It took eleven days from first commit to merge. One workflow mapped the correct Rust lifetime for every struct field. The next wrote each .rs file as a behavior-identical port. Hundreds of agents worked in parallel, with two reviewers per file. A fix loop then drove the build and test suite until clean. The result is not yet in production. What is Fast Mode Actually Fast mode is not a different model. It is a high-speed configuration of Claude Opus. It delivers 2.5x faster output token speeds. Model intelligence and capabilities stay the same. You toggle it with /fast in Claude Code. A small ↯ icon marks an active session. Fast mode trades cost for speed. For Opus 4.7 and 4.6, it is priced at $30/$150 per MTok. According to Anthropic official release, fast mode for Opus 4.8 is three times cheaper. It requires usage credits enabled on your account. It draws from those credits, not your plan’s included usage. It is best for rapid iteration and live debugging. Standard mode still fits long autonomous or cost-sensitive tasks. Strengths and What to Watch Strengths: Workflows handle codebase-wide bug hunts, large migrations, and cross-checked research end-to-end. Adversarial review and convergence aim for results a single pass cannot reach. Fast mode keeps Opus-level quality while cutting latency for interactive work. Resumable runs mean an interrupted workflow does not start over. What to watch: Both features consume meaningfully more tokens than a typical session. A run can spawn up to 1,000 agents, so costs climb fast. Fast mode bills from the first token at the higher rate. Both ship as research previews, so pricing and availability may change. Marktechpost’s Visual Explainer @import url('https://fonts.googleapis.com/css2?family=Newsreader:ital,wght@0,400;0,500;0,600;1,500&family=Hanken+Grotesk:wght@400;500;600;700&family=JetBrains+Mono:wght@400;500&display=swap'); #mtp-dwfm-guide{ --ink:#141413; --ink-soft:#39372f; --muted:#6b6862; --cream:#f0eee6; --paper:#faf9f5; --clay:#cc785c; --clay-deep:#b5613f; --line:#e4dfd2; --code-bg:#1f1e1d; --code-ink:#ece7da; background:var(--cream)!important; color:var(--ink)!important; font-family:'Hanken Grotesk',-apple-system,Helvetica,Arial,sans-serif!important; max-width:960px; margin:24px auto; padding:0; -webkit-font-smoothing:antialiased; box-sizing:border-box; border:1px solid var(--line)!important; border-radius:18px; overflow:hidden; } #mtp-dwfm-guide *,#mtp-dwfm-guide *::before,#mtp-dwfm-guide *::after{box-sizing:border-box;} #mtp-dwfm-guide hr,#mtp-dwfm-guide p:empty,#mtp-dwfm-guide del,#mtp-dwfm-guide s{display:none!important;} #mtp-dwfm-guide .mtp-bar{height:3px!important; width:100%; background:var(--line)!important;} #mtp-dwfm-guide .mtp-bar span{display:block; height:3px!important; width:10%; background:var(--clay)!important; transition:width .45s cubic-bezier(.4,0,.2,1);} #mtp-dwfm-guide .mtp-viewport{ display:flex; overflow-x:auto; overflow-y:hidden; scroll-snap-type:x mandatory; scroll-behavior:smooth; -webkit-overflow-scrolling:touch; background:var(--paper)!important; scrollbar-width:none; -ms-overflow-style:none;} #mtp-dwfm-guide .mtp-viewport::-webkit-scrollbar{display:none;} #mtp-dwfm-guide .mtp-slide{ flex:0 0 100%; min-width:100%; scroll-snap-align:start; padding:clamp(30px,5vw,60px); min-height:430px; display:flex; flex-direction:column; justify-content:center;} #mtp-dwfm-guide .mtp-kick{font-family:'JetBrains Mono',monospace!important; font-size:11px; letter-spacing:.22em; text-transform:uppercase; color:var(--clay-deep)!important; margin:0 0 18px;} #mtp-dwfm-guide h2.mtp-title{font-family:'Newsreader',Georgia,'Times New Roman',serif!important; font-weight:500; font-size:clamp(26px,4.4vw,40px); line-height:1.12; letter-spacing:-.01em; color:var(--ink)!important; margin:0 0 16px;} #mtp-dwfm-guide .mtp-sub{font-size:clamp(15px,2vw,18px); line-height:1.55; color:var(--ink-soft)!important; margin:0; max-width:46ch;} #mtp-dwfm-guide ul.mtp-list{list-style:none; margin:6px 0 0; padding:0;} #mtp-dwfm-guide ul.mtp-list li{position:relative; padding:0 0 0 26px; margin:0 0 14px; font-size:clamp(14.5px,1.9vw,17px); line-height:1.5; color:var(--ink-soft)!important;} #mtp-dwfm-guide ul.mtp-list li::before{content:""; position:absolute; left:0; top:.62em; width:9px; height:9px; border-radius:50%; background:transparent!important; border:1.5px solid var(--clay)!important;} #mtp-dwfm-guide ul.mtp-list li b{color:var(--ink)!important; font-weight:600;} #mtp-dwfm-guide code{font-family:'JetBrains Mono',monospace!important; font-size:.88em; background:rgba(204,120,92,.12)!important; color:var(--clay-deep)!important; padding:1px 6px; border-radius:5px;} #mtp-dwfm-guide pre{background:var(--code-bg)!important; color:var(--code-ink)!important; border:1px solid #2c2a27!important; border-radius:10px; padding:18px 20px; margin:18px 0 0; overflow-x:auto; font-family:'JetBrains Mono',monospace!important; font-size:13px; line-height:1.7;} #mtp-dwfm-guide pre code{background:transparent!important; color:var(--code-ink)!important; padding:0; border:none!important;} #mtp-dwfm-guide pre .cmt{color:#8a8579!important;} #mtp-dwfm-guide pre .cmd{color:#e0a07f!important;} #mtp-dwfm-guide .mtp-stat{display:flex; flex-wrap:wrap; gap:10px 30px; margin:8px 0 6px;} #mtp-dwfm-guide .mtp-stat .num{font-family:'Newsreader',Georgia,serif!important; font-weight:600; font-size:clamp(24px,3.6vw,34px); color:var(--clay-deep)!important; line-height:1; display:block;} #mtp-dwfm-guide .mtp-stat .lab{font-size:12.5px; color:var(--muted)!important; margin-top:6px; display:block;} #mtp-dwfm-guide .mtp-mark{font-family:'JetBrains Mono',monospace!im